UNIVERSITY OF PANNONIA GEORGIKON FACULTY DOCTORAL SCHOOL OF ANIMAL AND AGRO-ENVIRONMENTAL SCIENCES Head of Doctoral School: Dr. habil. Angéla Anda, Professor Supervisor: Dr. habil. Katalin Sárdi, Professor Co-Supervisor: Dr. József Berke, Associate Professor APPLICATION OF DIGITAL IMAGE PROCESSING IN THE EVALUATION OF GREENHOUSE AND FIELD EXPERIMENTS THESES OF DOCTORAL (PhD) DISSERTATION Written by: GERGELY GRÓSZ Keszthely 2010
1. BACKGROUND, SUBJECT MATTER, GOALS AND CURRENT VALIDITY OF THE RESEARCH Recently, Information Technology plays an important role in agricultural sciences, similarly to other disciplines. Rapidity has become one of the most important factors in evaluating results of scientific experiments. Data information may be sent from the field with a mobile phone or wireless Internet connection to the laboratory staff for assessment and interpretation. With this approach, using specific equipments the costs of travelling and experiment can be reduced. In lots of cases the colour information examinations, the experiments are evaluated with visual techniques, which are often full with subjective failures. To except these failures one possibility is to process objective computer based examination methods. The use of information technology increases, which coheres, that the digital images can be found anytime, so the examinations can be repeated with new technology, so the results can be compared to each other. The other important reason of the increasing application of digital technology is that the prices decrease constantly by better image quality. Objectives The main objectives of my PhD doctoral research were to correct and (to) work out the methods, to process the digital images, taken with a digital camera in the visual and infra spectrum at application in agricultural sciences: Comparing analogous and digital methods at variable nutrient supply levels using sunflower, corn and potato test plants; Process together multispectral and multitemporal information by nutrient supply; Spectral and structural analysis with image processing classification, comparing different visual sensors at digital leaf area metering; Comparative analysis of the results with the local diagnostic methods; Developing a model of a decision helping system from the results. 2
2. MATERIALS AND METHODS Leaf area and colour information experiment, using 20 test plants The following crops were examined: winter wheat (Triticum aestivum ssp. vulgare), rye (Secale cereale L.), spring barley (Hordeum vulgare L.), oats (Avena sativa L.), corn (Zea mays L.), sugar-beet (Beta vulgaris L. var. altissima Doell.), potato (Solanum tuberosum), green peas (Pisum sativum L.), soybean (Glycine max L. Merrill), beans (Phaseolus vulgaris L.), chick-pea (Lathyrus cicera), sunflower (Helianthus annuus L.), rapeseed (Brassica napus L. ssp. oleifera), flax (Linum utitatissimum L.), alfalfa (Medicago sativa L.), english clover (Trifolium pratense L.), tomato (Lycopersicon lycopersicum L. Karsten ex. Farwell.), bell pepper (Capsicum annuum L.), carrot (Daucus carota L.), and grape (Vitis vinifera). The research was carried out in Mitcherlich pots using 4 kilograms of a mixture of a calcareous chernozem soil and mould with a ratio 3:1. Parameters measured at sampling were as follows: upper and lower leaf blade colour examination, in visible and infra red spectrum, leaf area measurement, using squared-plotting paper, digital camera, scanner and mobile phone. Nutrient supply experiment under greenhouse conditions The experiment was carried out under greenhouse conditions using sunflower and corn test plants in 2006 and 2007 on calcareous chernozem soil in 6 kilograms Mitcherlich pots with each 8-8 plants in each year. The amounts of nutrients applied in the treatments: N1 = 80 mg/kg, P1 = 40 mg/kg P 2 O 5, K1 = 120 mg/kg K 2 O. 6 treatments were applied throughout the experiment: N0P0K0, N1P1K1, N1P1K0, N2P2K0, N0P0K1 and N0P0K2. Samplings were taken at the growth stage of 6-8 leaves for sunflower, at the same growth stage and at bloom for corn, taking 2-2 samples from each plot. Nutrient supply experiment under field conditions The nutrient supply experiment was carried out in 2006 and 2007 using sunflower (30 m 2 plots), corn (30 m 2 plots) and potato (47 m 2 plots) as test plans. The sunflower and corn were grown in a calcareous chernozem soil and potato was grown in a Ramann type brown forest ground. 3
The amounts of nutrients applied in the treatments are given (2006, 2007): sunflower (2006, 2007): N 70 kg/ha P 2 O 5 168 kg/ha K 2 O 287 kg/ha; corn (2006, 2007): N 240 kg/ha, P 2 O 5 60 kg/ha, K 2 O 372 kg/ha; potato (2006): N 112,5 kg/ha, P 2 O 5 62,5 kg/ha, K 2 O 250 kg/ha; potato (2007): N 105 kg/ha, P 2 O 5 77,5 kg/ha, K 2 O 237,5 kg/ha. 6 treatments were applied throughout the experiment: N0P0K0, N1P1K1, N1P1K0, N2P2K0, N0P0K1 and N0P0K2. Samplings were taken 3 times at the sunflower and corn at age of 6-8 leaves, blooming and harvest. Samplings were taken 2 times at the potato at age of blooming and harvest. Parameters measured were the following: median of histograms of leaf blade colour by treatments, leaf fresh and dry matter weight (g/plant), stem fresh and dry matter weight (g/plant), flower fresh and dry matter weight (g/plant), yield fresh and dry matter weight (g/plant), leaf area (cm 2 /plant), N total %, P 2 O 5 %, K 2 O% of samples Parameters of leaf area measurement methods comparison In this experiment analogue squared-plotting paper, LI-COR LI-3000A, Hewlett-Packard HP4670c scanner, Canon EOS 10D and 1D Mark III digital cameras, Sony Ericsson K750i mobile phone and a Hexium VariCam Infra Red Camera were used. The images were analysed using Adobe Photoshop CS3 and Gimp 2.6 software. Statistical analyses (ANOVA, i.e. analysis of variance and correlation calculation) were made with using Microsoft Excel 2003 computer program package. Description of the digital leaf area measurement At first reference images (millimetre scale on the sides of the images) should be opened, wherewith the area of the pixels was calculated if manufacturer s standards are not available. In the next step the images containing the leaves were opened with the software. The leaves were cut around as closely as possible to avoid mistakes. Then Median filter was applied thus point-like noise were reduced. Threshold function was selected thus the object was separated from the background. After Threshold function the pixels not wholly covered was corrected which appeared mostly at the protuberances of the veins. The object pixels were 4
counted with the help of the Histogram option and they were multiplied by the pixel area thus the area of the leaves was obtained in cm 2. 5
3. RESULTS AND CONCLUSIONS Results of plant leaf comparisons In this experiments, leaf colour of 20 test plant species were compared in visible and infrared spectrum. From the results, it was found that the leaves of the plants could be separated very clearly in visible range of the spectrum at the same conditions. Certainly, soybean and rapeseed could be separated from the others, their median of histograms were significantly different from the other plants. Rye, oats, corn, green peas, chick-pea and the english clover showed the worst rate. In that case, the median was only 47.36% statistically different. To summarise from the 190 test plant pairs 130 was statistically different which was 68.42%. With more correction of this method, the results could be much better and used at other images taken with remote sensing in visible spectrum. In infrared spectrum the leaves could be separated less easily than in visible spectrum. To summarise from the 190 test plant pairs 117 was statistically different which was 61.58%. In this spectrum, sugar-beet and sunflower leaves gave the best results. Rye and spring barley, corn, green peas, alfalfa, the english clover, bell pepper and carrot samples showed lower rates which was 52.63%. It s suitable to compare the results in other examination with aerial and satellite photos, however, the effects of intensity modification of the atmosphere should be considered. Evaluation of the K nutrient supply pot experiment with corn using histogram of digital images To compare upper and lower leaf blade of corn test plant, the median of histograms of RGB and green channels of images which contains the upper and lower leaf blade at the growth stage of 6-8 leaves and blooming were analysed. From the two values at both sampling at the case of green and RGB channels a proportional number was made and the maximum, the minimum value, average difference and standard deviation were counted. From the comparison of analysis results I found that there were close relationships between the histograms of images which contained upper and lower leaf blade at both sampling dates. At the growth stage of 6-8 leaves, the correlation coefficient was R 2 =0.7713 at the 3 channel, and in the green channel it was R 2 =0.7441. The values of different areas at the growth stage of 6-8 leaves fluctuated very much but the ratio to each other was found to be homogeneous. To compare to the first sampling period, the correlation was closer at blooming. At the 3 channel 6
the correlation coefficient was R 2 =0.9260, and in the green channel it was R 2 =0.8809. At blooming the ratio between the upper and lower side of the leaf was more exact by corn test plant. The results of the examination could be used by a decision helping system, however, it was necessary to enlarge the date with other crop species and varieties. To evaluate the potassium nutrient supply experiment with the digital method, the images were taken from upper and lower leaf blade of corn test plant grown at different potassium nutrient supply levels. The histogram of green channel was higher than the same parameter of RGB channels with 23-24%. The tendency was similar at the RGB and green channels. At the first sampling time the lower leaf blade, 40% of the data pairs were statistically different at RGB channel and it was 33% by the green channel, At the colours these rates were 46.7% and 53.3%, respectively. At the second sampling date, the significant data pairs amounted 60% in every examination. At the first sampling the 002 treatment can be separated well in every channels and every parameters. At the second sampling date, the 110 and 220 treatments were significantly different from the others. This method did not show any significant differences in the other case, but the efficiency of the method could be corrected with stabilization of the conditions. Application of other nutrient elements, e.g. nitrogen could be studied for the wider understanding of the relationship between nitrogen supply level of the plants and the intensity of the colour. The setting up of the median based decision helping system The system includes three modules: In the first part the upper and lower leaf surface can be compared with the healthy, the unhealthy and the leaves with nutrient deficiency results can be stored in a database. When these are significantly different, the software sends a failure message. Above some failure messages it is purposed to make analytical diagnosis. The results of the images can be stored in the database, the comparison base grows, so the examinations will be more correct. Another database contains similar parameters of a nutrient supply experiment, which can be loaded and compared to the database. The examination can be done with a well defined colour chart; that is used successfully in practise for rice crop. By the digital methods, the colour chart was scanned before the examination because the environmental parameters were the same when using the same scanner. Nitrogen levels can be deduced from the chart which was shown in my research. The third module is an automatic leaf area, leaf length and width measurement application, which calculates the leaf length and width with a well defined area. 7
The best feature of the application is enlarging and handling of data, and an Internet based database where the users can update their colour database. Results of the leaf area measurement methods For the calibration, exactly defined square areas and for the examination of 20 test plant species were used. The effect of the angle position of digital camera on the result of leaf area measurements Unit (100 cm 2 ) area was placed over the work-table. The camera was placed on a stand and rotated from 90 to 30 by 5. Four images were taken in every position. The data were measured by using the Photoshop program. It was established that the rotation angle was decreased, the leaf area was changed significantly, and its volume became larger with the rotation angle. The effect of sensor sensitivity on the result of leaf area measurements In this study, the unit (9 cm 2, 25 cm 2, 100 cm 2 ) areas were photographed with different ISO (100-3200) sensitivity in four replicates. From the results it was evident that taking photos at different ISO sensitivity did not cause significant. The effect of the lenses applied on the result of leaf area measurements In this experiment the effect of different lenses on the result of leaf area measurements were evaluated at the same aperture value (f/11). The following lenses were connected to the Canon EOS 10D body: Sigma DL Zoom 35-80mm f/4-5.6 at 35mm, 50 mm and 80 mm, Canon EF 50 mm f/1.8 II at 50mm, Sigma AF 105 mm f/2.8 EX Macro DG Macro at 105 mm. The largest difference from the real value was found at the Sigma DL Zoom 35-80 mm f/4-5.6 at 35 mm: 2.71%. At the others there were not significant differences. The distortion caused the big difference. The effect of aperture values on the result of leaf area measurements The effect of aperture values were examined with an aperture line from 1.8 to 22 using a Canon EF 50 mm f 1,8 II lens on a Canon EOS 1D Mark III body. The flaw was found from 8
0.91% to 1.65%. Significant differences were found at some results. The differences were found because the lenses mapping. Possibility of using mobile phones Photos were taken with 3 resolutions (160*120, 640*480 and 1632*1224) with a Sony Ericsson K750i mobile phone. Reducing the resolution the difference is becoming larger. The exact indication of squared-plotting paper in the smallest resolution could be a problem. At the largest resolution the difference comes from the wide angle (28 mm) distortion. Comparing the results with a digital camera unit having bigger sensor and better quality lenses it could be seen that the error is almost double. The effect of different devices on the result of leaf area measurements The following methods were used for the comparison: squared-plotting paper (as a reference), LI-COR LI-3000A (table and portable) analogue leaf area meter, Hewlett-Packard HP 4670c scanner and Canon EOS 1D Mark III body with Canon EF 50 f/1,8 II lens. To process images Adobe Photoshop CS3 professional image processing software was used. For the comparison 9 cm 2, 25 cm 2, 100 cm 2 squared-plotting paper as well as monocotyledons and dicotyledonous leaves were used. The comparison was made with the test plants too. Leaves of several plant species could not be measured with certain methodologies with the squared-plotting paper the carrot and with the Li-3000A mobile version the chick-pea and flax. Considering the squared-plotting papers mistake (1.5%) the mobile leaf area measurer gave the most inexact result considering both the test areas and plants due to the usage conditions (Figure 1-2). 9
Difference (%) Difference (%) 14 12 10 8 6 4 2 0-2 LI-3000A (table) LI-3000A hp 4670 c (handy) Methods Canon EOS 1D+EF 50 Average difference Standard deviation (n=50) Figure 1: Comparison of results of leaf area measurement methods using reference areas 9 cm 2, 25 cm 2, 100 cm 2 squared-plotting paper, a monocotyledonous and a dicotyledonous leaves with mean and standard deviation (n=50) 25 20 15 10 5 0-5 Average difference Standard deviation -10 Li-3000A (table) Li-3000A hp 4670c (handy) Methods Canon EOS 1D+EF 50 (n=160) Figure 2: Comparison of results of leaf area measurement methods using test plants with mean and standard deviation (n=160) The digital scanner method gave the most exact result in both cases considering the standard deviation too. Other advantage of the method is pressing the objects. Biggest standard deviation was obtained by the different press approaches and the variable leaf morphology. The most important aim of the experiment was to find a low cost method, so the most exact digital scanner method was analysed with free Gimp 2.6 image processor. In that case the 10
squared-plotting paper was the basic as well, to that was calculate average difference and standard deviation. The average difference was 0.55%, while the standard deviation was 0.66%. Summarizing the results it can be stated that digital leaf area measurement can replace the analogue leaf area meters. Comparison of leaf area measurement methods with other approaches To compare the digital methods with the estimation approaches it can be stated that those methods use different proportional numbers in every case in front of the maximal leaf height and width product. The method presented can be used with the same parameters for every plant species. Beerling and Fry used microprocessor methods, at the smallest leaves the difference compare to the print method were 10.46% by the largest leaves 1.28%, but they found their method faster. On the other hand, best values of the tested methods were obtained using a scanner, 2.43% average difference was observed from the print method. The cost of my methods was compared to the WinFOLIA, which costs from 265 thousand HUF (Hungarian Forints) to 932 thousand HUF, depending on the modules. When the measurement was made with the Gimp 2.6 instead of Photoshop the software did not have any costs. The cost of the leaf area measurement methods Prices were given at December 2009. The costs of hardware were about 86 thousand HUF (personal computer 70 thousand HUF, scanner or digital camera 15 thousand HUF). The prices of the used software were the following: Microsoft Windows XP Home 23 thousand HUF, Microsoft Office 2007 Home and Student 17 thousand HUF, Adobe Photoshop CS3 203 thousand HUF. The Linux operating system and Gimp image processor can be downloaded from the Internet for free. The AM 100 portable leaf area meter made by ADC BioScientific Ltd costs approximately 1,3 million HUF. In this case to fix the data, a computer is also required. To consider all parameters and costs the best method is using a scanner and Gimp 2.6 software for evaluation. Results of the nutrient supply experiments The digital leaf area measuring method was used successfully by 3 test plants (sunflower, corn and potato) to evaluate potassium nutrient supply experiments under greenhouse and field conditions. 11
Results of sunflower potassium nutrient supply experiment under greenhouse conditions There were significant differences of every treatment in case of average height; the optimal nutrient supply treatment (111) gave highest values. Out of the control in every treatment the 2007 results were higher on the average of 26.13%. In the case of leaf area, there was a significant difference compared to the control in both experimental years. In 2006 and 2007, the leaf area of plants receiving the optimum nutrient level (treatment code: 111) was the biggest. The tendency in green mass production (fresh weight) of the leaves was similar to that of the leaf area. In 2006 and 2007, the leaf area of the plants receiving 111 treatment was the biggest. There was a significant difference between most of the treatments and the unfertilized control plants only in 2007. Average dry weight of leaves showed significant differences compared to the control. At higher potassium nutrient supply, the water content in several treatments was lower than in samples taken in 2006. The difference from control in fresh weight was significant in each experimental year. Similar to the other parameters, the effect of the optimum 111 treatment was the most favourable. The tendency of dry matter production of stems showed significant differences between the control plants and the other treatments. A significant difference was found compared to the control in every case by N total %, the highest were for the 111 and 110 treatments. Most of the effects of treatments were statistically different. In the case of P content, (P 2 O 5 %) the highest values were obtained in the 220 treatment in both experimental years. At the stems in the first year the 001 in the second the 110 treatment gave the highest value. Except the 111 and 110 treatments significant difference was found between the control and the treatments. In the case of K 2 O% it can be observed in both two years by leaves and stems that the 001 and 002 treatment gave the highest value. In the case of harmonious nutrient supply (111) the values were high too. Significant difference was found when compared to the unfertilized control at most of the treatments. Results of corn potassium nutrient supply experiment in greenhouse Compared to the control in each experimental year, at both sampling date, significant differences were found in every treatments except to the 110 and 220 as well as 001 and 002 treatments in values of average plant height. 12
The tendency of the average leaf area was the same in both years. There were significant differences compared to the control except the second sampling date for the 001 and 002 treatments. The green mass of the leaves was highest at the treatment 111, 110 and 220 in both sampling date, N level was the highest. There were significant differences compared to the control in the average dry matter weight in both years. Except to the control, treatments were square at the stem fresh weight, because of the effect of the potassium. The dry matter of stem was not statistically different in every case, except to the first sampling in 2006. The tendencies were similar at each sampling date. To evaluate N total % it was suggested that the leaves and stems gave the best results at the optimum and at the relative excess of N. To evaluate P content, (P 2 O 5 %) it was evident that the control gave the lowest values. The highest values were obtained by the optimum and relative excess of phosphorus treatments. The level of K 2 O% in both experimental year was higher at the stage of 6-8 leaves in leaf and stem samples. Results of sunflower potassium nutrient supply field experiment In every 3 sampling in both experimental years there were significant differences in average plant height compared to the control in each treatment. In both sampling date, plants were higher at the optimal and the relative excess of potassium treatments. Evaluating leaf area results, significant differences were found compared to the control and among treatments, except 110 treatment at first sampling in 2006. In every case the tendency was similar except the first sampling. The highest leaf area was given by treatment 110. The control had the smallest area in every sampling. Evaluating the fresh weight of leaves it can be stated, that the II sampling in 2007 excels, because of better environmental parameters. Such as the leaf area, the 110 treatment gave the biggest value. Compared to the control all of the treatments in every sampling were higher, and this difference was significant in the most cases. At the dry matter weight the tendency was the same as the fresh weight. The differences were similar to the treatments, but most of the treatments it wasn t significant compared to the control or each other. In the fresh weight of stems the sampling at blooming in 2007 excels too, but it wasn t so big than in the same time in 2006. At the first sampling, differences were statistically proved when compared to the control but only the tendency was similar at blooming and harvest, differences were not significant. To evaluate the dry matter weights it can be stated that in 13
most of the treatments there were significant differences compared to the control. In both years the weight at blooming was higher than that of at harvest except 3 treatments in 2006. The average fresh and dry matter weight of blooms was larger in 2006 unlike the weight of the leaves and stems. There was a significant difference in half of the treatments compared to the control at fresh and dry matter weight too. Above the weights, the diameter was evaluated too. At this parameter, values obtained in 2006 were higher. Significant differences were observed for several treatments compared to the control. It can be stated that at this parameter 111 (the optimal) and the 110 (nitrogen excess) gave the best result. To evaluate weight of seeds it turned out, that the best result was given by 111 optimal nutrient supply treatments at except 001 in 2007 at the fresh and dry matter weight. Significant differences were obtained in both years compared to the control. The average diameter showed statistically difference compared to the control. In 2007 the tendency was similar but the difference wasn t significant. The N total % of the leaves was highest at treatment 110 and 220 at each sampling date in 2006 and 2007. This result was statistically different in most of the cases. To evaluate the stems the treatment 111, 110 and 220 gave high value in every sampling in both years. By the blooms and seeds the tendency was similar and the values were significantly different. The values of 2007 were higher than the 2006 s, but it was the opposite for the values of seeds. To evaluate P 2 O 5 % at the leaves it was stated that treatment 111 gave the highest result. Significant differences were found compared to the control for most of the treatments. Treatments which received higher potassium levels gave smaller values because of the one sided nutrient supply in every years. At the stems it gave the similar results as it was found for the leaves. Best result were obtained at treatment 111, differences were significant where compared to the control most of the cases. At the blooms the tendency was different, because in the first year the 110, in the second the 220 treatment gave the highest value. Except the 001 in 2006 and the 002 in 2007 significant difference was found compared to the control. By the P 2 O 5 % at stems the treatment 111 showed a high value in both two years. At this parameter the values weren t significantly different. The average K 2 O% at the leaves was the highest by the optimal and potassium excess treatments. Significant differences were observed when compared to the control. The tendency at the stems was the same as at the leaves. At this parameter, the ratio of the statistically different treatments increased. At the blooms and seeds the tendency was similar to the P 2 O 5 %, lower values were shown at the blooms, opposite the seeds. 14
Results of corn potassium nutrient supply experiment in field In the potassium nutrient supply experiment carried out under field conditions, no significant differences were found in most of the treatments by plant height. The 2007 data were higher than in 2006 in every sampling. On the other hand, significant differences could be observed for about half of the treatments by the leaf area. The tendency was similar all of the samplings, best results were given by the optimum and potassium excess treatments. Similarly to average values in leaf area, fresh and dry matter weight of the leaves were higher in 2006. No significant differences were found when compared to the control and among the treatments. Such as the leaves, I didn t found significant differences in most of the treatments at the fresh weight of the stems. At the dry matter weight of stems the results were more balanced between the II and III sampling. More significant difference were found between the treatments than at the fresh weight. The fresh and dry matter weight of bloom in 2006 exceeded significantly the weights in 2007. It was found that most of the treatments statistically differed compared to the control. The optimal treatment gave the highest value in both years. At weight of seeds and the length of the corncobs I only found significant difference at treatment 110 and 220 significant difference. Compared to the control longer corncob and larger yield weight was measured. The optimal treatment (111) gave the highest level too. At the leaves the total nitrogen level at treatment 111 and 110 was the highest. On the other hand, treatments showed the deficit of other essential nutrients: values of treatment 002 were lower than those of the control. To evaluate the stems it was suggested that values of the first sampling exceeded those of the second and third sampling date. Best results were given by optimum and by the relative nitrogen excess treatment. To evaluate the blooms it was stated that treatment 002 give the best result in both experimental years. Highest values were obtained for seeds at the treatment 111 and 110 in both years, significantly exceeding the value of the control. At the P 2 O 5 % at leaves the highest value was given by treatment 111, 110 and 220. Gradual decreases were found at the other treatments. The 111 and 110 treatments varied statistically compared to the control. The first sampling exceeds in both two years. It can be stated that at each sampling dates, the values of 2006 were higher than those in 2007. At the blooms the values were higher in 2006 as well. At the seeds the difference between of two treatments was much higher. 15
At the leaves the average K 2 O% values was higher except treatment 000 in 2006 than in 2007. It was found that treatment 111 gave the best result at most of the samplings. The values of the first sampling was exceed in both years. It was found that the values of the first sampling exceeded in both years at the stems too. The difference at the leaves and stems were significant compared to the control. Except treatment 111 the values in 2006 were higher than in 2007 in the blooms. There were only few cases where I found significant differences compared to the control. To evaluate the seeds it was evident that the values in 2006 were higher too. In both experimental years treatments with relative excess of potassium showed higher results. Results of potato potassium nutrient supply field experiment There were significant differences of treatments of plant height compared to the control in 2007. It was only two times (110, 001) in 2006. The tendency of the treatments was similar in both years. To evaluate the leaf area treatment 220 exceeded the control in 2006, however, it was different in 2007. The highest result was given by treatment 111 (balanced nutrient supply) in both years. While there was a significant difference compared to the control in the first year, then in 2007 it wasn t significant in every cases. Average fresh and dry weight of leaves showed similar tendency in both two years. Differences were significant compare to the control in both years. The optimum treatment (111) in 2006, and in 2007, the treatment 002 (relative excess of potassium) showed highest values. Significant differences were found only for average dry weight. Average fresh and dry weight of stems showed similar changes to those of weight in the leaves. The highest value was shown by the optimal and the treatment of relative excess of potassium. The tendency of yield was the same in both years. Examining harvest quantity, however, significant results were obtained compare to the control. The treatment 111 was higher than the control in 2006 at 46%, till in 2007 at 36%. There were no significant differences between the treatments in the case of dry matter % except treatment 110 in 2007. The tendency was the same among to the treatments in both years. There were significant differences of treatment also in case of N total % in 2006, but in 2007 it was only 3 times. The optimum treatments gave the best results in both years. There were only a few treatments which differed significantly at the stems compared to the control in 16
2006. Except to treatment 001 in every case the difference was significant in 2007. The tendency at the tubes showed a slight difference because the treatment 001 and 002 gave the highest result. Differences were significant only in 2007 in few cases. To evaluate the P 2 O 5 % it could be stated likewise the other optimum treatments there were significant differences compare to the control at the leaves in both years. Significant differences were observed among the treatments in 2006. The highest P 2 O 5 % level was at treatment 111 and 220. Differences were similar to the leaves and stems compared to the control and among to the treatments. Treatment 111 and 220 were at the highest values. To evaluate K 2 O%, significant differences were obtained when compared to the control for the leaves. Differences were significant and similar for the stems in both experimental years. It was found that significant difference was only in 2006 at the tubes at the K 2 O%. 17
4. CONCLUSIONS Nowadays the information technology and its toolkits have become am important part of the agricultural production and research. More and new information can be obtained by application of these, which are more correct, faster and we can make objective adjudication. The subject of my PhD research was to specify and work out usable methods in research of crop production. To reach that, experiments series were carried out. During my studies, nutrient supply experiments were conducted with sunflower, corn and potato in 2006 and 2007, and in 2008 to compare the leaf area measurement methods with 20 test plant species. In this experiments, leaf colour values were compared in visible and infrared spectrum with the test plants. For the image processing Adobe Photoshop C3 for the statistical analysis (analysis of variance, correlation analysis) Microsoft Excel 2003 were used. The sunflower and corn experiments under field conditions were set in 30 m 2 plots while potato was grown in 47m 2 plots. Sunflower and corn experiments were carried out in a calcareous chernozem soil and the potato experiment was carried out on a Ramann type brown forest soil. Samples were taken 3 times at the sunflower and corn at the growth stage of 6-8 leaves, blooming and harvest at the potato at age of blooming and harvest. 6 treatments were applied throughout the experiment. Parameters measured were the following: leaf fresh and dry matter weight, stem fresh and dry matter weight, flower fresh and dry matter weight, yield fresh and dry matter weight, leaf area, N all %, P 2 O 5 %, K 2 O%.To compare the different nutrient supply treatments and the upper and lower leaf blade the images which contain the corn leaves at greenhouse conditions were used. From the results, a model of a decision helping system was developed. To summarise the separation of different test plant to valuate the images histograms it was stated that 68.42% was statistically different in visible spectrum, while in infrared spectrum it was 61.58%. Certainly the soybean and the rape can be separated from the others, those were 100% significantly different to the other plants. The results in infrared spectrum were all of the cases much wrong. With the potassium nutrient supply experiment, several other studies were conducted with corn as a test plant. Upper and lower leaf blade were compared using the median of RGB and 18
histograms of green channel of images which contains leaf blades at age 6-8 leaves and blooming. From the two values at both sampling date at the case of green and RGB channels a proportional number was made and the maximum value, the minimum value, average difference and standard deviation were counted. I found that there were close relationships between the histograms of images which contained upper and lower leaf blade the correlation coefficient was between R 2 =0.7441 and R 2 =0.9260. At blooming, the ratio between the upper and lower side of the leaf was more exact by corn as a test plant. The results of the examination could be used by a decision helping system, but it was considered necessary to complete the database with other crop species and varieties. Besides the comparison it was observed that for leaves having different nutrient supply, treatments could be separated at 30-50% accuracy. The best perceptible treatment was the 220 (relative excess of nitrogen and phosphorus) which separation was significant in most of the cases. The method would be analysed better in case of nitrogen nutrient supply experiments because of the correlation of the leaf colour and nitrogen, treatment 220 showed itself. The decision helping system consists of 3 modules: Examination of the ratio of upper and lower leaf surfaces, Examination of nutrient supply comparison and Leaf area measurement module. It can be stated that the rotation angle of the camera is critical at leaf area measurement. It was established that the rotation angle was decreased by 5, the leaf area was changed significantly compare to the control. The sensitivity of the sensor caused significant aberration only over ISO 1250 and above. The quality of the lenses has an effect of leaf area results because the distortion changes the values. The effect of flaws of aperture values was found from 0.91% to 1.65% which was found because of the mapping of the lenses. Using mobile phone it was stated that the 640*480 resolution didn t influenced significantly to the results. To compare the devices it was observed that in the most inexact method was the handy leaf area meter, and the most exact was the digital method using scanner which was the most expend effect, too. To compare with the earlier used methods the digital leaf area measurement method is faster, it was not species-specific and cheaper. 19
Depending on the quantity of the used fertilizers at the nutrient supply experiments, the vegetative and nutrient analysis parameters of plants changed significantly. The best results were given at most of the parameters by treatment 111 (optimal nutrient supply). With this result it turned out, that the experiments were successful and the nutrient quantities were suitable. 20
5. NOVEL SCIENTIFIC RESULTS 1. During my research, it was established that leaves of the plants could be separated very clearly using the median of histograms of digital images taken in visible range of the spectrum. The accuracy of examination depends on plant species and environmental conditions. The precision of results was less correct in the infra red spectrum range in each case. 2. Close relationships were found between the upper and lower leaf blade of corn test plant using the median of histograms of digital images in visible and infra red spectrum as well. Further studies are needed using other crop species as test plants. 3. Nutrient supply treatments could be separated with corn test plant using the median of histograms of digital images in the visible spectrum range. The most accurate results were obtained with the highest rate of nitrogen supply indicating the positive, synergistic relationship between the two nutrient elements. For the wider understanding of these interrelations, results of further experiments would provide essential and detailed information on the effects of nitrogen supply. 4. Using regular image processor software with a scanner, digital leaf area measurement methods could replace the analogue leaf area meters in crop production experiments at low cost level as it is more accurate compared to conventional methods. 5. The methods of measurements which were used in the experiments may serve as a good basis for developing a decision-making system which helps the user to evaluate plant health, nutrient-supply and leaf area parameters of the examination and in solving the problem. 21
6. FURTHER RESEARCH DEVELOPTMENT Differences in health conditions related to nutrient supply, areas showed differences at lightning distribution function. The necrotic areas were much different in histogram than in the healthy ones. To make a reference database, at same environmental conditions we could deduce to the nutrient deficiencies, or illness. Further research is needed for evaluating colour information obtained in other nutrient supply experiments with other macro-nutrient elements. 22
7. LIST OF RELATED PUBLICATIONS PROCEEDINGS PUBLISHED IN HUNGARIAN LANGUAGE CONFERENCE TRANSACTIONS GRÓSZ, G. BALIKÓ, K. (2006): Fiatalkorú borsó (Pisum sativum L.) és szója (Glycine max L. Merrill) vegetatív paramétereinek alakulása a tápanyag-ellátás függvényében. 7. RODOSZ Tudományos Konferencia, Kolozsvár. pp. 40-41. GRÓSZ, G. BALIKÓ, K. (2006): A digitális levélfelület-mérés gyakorlati alkalmazása borsónál és szójánál. XII. Ifjúsági Tudományos Fórum, Pannon Egyetem Georgikon Mezőgazdaságtudományi Kar, Keszthely. pp. 1-5. GRÓSZ, G. (2007): Burgonya (Solanum tuberosum L.) kálium tápanyag-ellátási kísérlet kiértékelése különböző módszerekkel. Képfeldolgozók és Alakfelismerők VI. konferenciája, KÉPAF 2007, Debrecen. pp. 104-112. GRÓSZ, G. (2007): Üvegházi napraforgó (Helianthus annuus L.) kálium tápanyag-ellátási kísérlet kiértékelése különböző módszerekkel. XIII. Ifjúsági Tudományos Fórum, Pannon Egyetem Georgikon Mezőgazdaságtudományi Kar, Keszthely. CD ROM pp. 1-5. GRÓSZ, G. SÁRDI, K. BERKE, J. (2007): Elektronikus tananyag az agrokémia tantárgyhoz az agrár-felsőoktatásban. Multimédia az Oktatásban Konferencia, Budapesti Műszaki Főiskola, 2007. augusztus 23-24. GRÓSZ, G. (2008): Üvegházi napraforgó kísérlet eredményeinek bemutatása két év alapján, digitális és analóg mutatók segítségével. XIV. Ifjúsági Tudományos Fórum, Pannon Egyetem Georgikon Mezőgazdaságtudományi Kar, Keszthely. CD ROM pp. 1-6. GRÓSZ, G. (2008): Burgonya (Solanum tuberosum L.) kálium tápanyag-ellátási kísérlet kiértékelése különböző módszerekkel két év adatai alapján. Informatika a Felsőoktatásban 2008, Debreceni Egyetem, Informatikai Kar, Debrecen. pp. 37. 23
GRÓSZ, G. SÁRDI, K. BERKE, J. HEGEDŰS, G. (2008): Elektronikus agrokémia tananyag az agrár-felsőoktatásban. Informatika a Felsőoktatásban 2008. Debreceni Egyetem, Informatikai Kar, Debrecen. pp. 83. PROCEEDINGS PUBLISHED IN FOREIGN LANGUAGE CONFERENCE TRANSACTIONS BALIKÓ, K. GRÓSZ, G. SÁRDI, K. (2006): Changes within in the leaf area of pea (The Pisum Sativum L.) as affected by nutrient supply and the age of plants. European Union 3 RD International Conference, Mosonmagyaróvár, Hungary. pp. 85-92. GRÓSZ, G. SÁRDI, K. BERKE, J. (2007): Evaluation of an experiment on the potassium nutrient supply of potatoes (Solanum tuberosum L.). Internatonal Conference on Agricultural Economics, Rural development and Informatics, Debrecen. pp. 315-324. ARTICLES PUBLISHED IN HUNGARIAN LANGUAGE JOURNALS GRÓSZ, G. SÁRDI, K. BERKE, J. (2007): Napraforgó (Helianthus Annus L.) kálium tápanyag-ellátási kísérlet eredményei. Acta Agronomica Óvariensis Vol. 49, No. 2 I. kötet, Competitor 21 Kiadó. pp. 345-352. GRÓSZ, G. SÁRDI, K. BERKE, J. (2007): Elektronikus tananyag az agrokémia tantárgyhoz az agrár-felsőoktatásban. Alkalmazott Multimédia 3./II./2007 Apple Magyarországi Képviselet, HDSYS Kft. pp. 87-91. ARTICLES PUBLISHED IN FOREIGN LANGUAGE JOURNALS GRÓSZ, G. SÁRDI, K. BERKE, J. (2007): Electronic textbook for agrochemistry in agricultural higher education. Journal of Applied Multimedia 3./II./2007 Apple Hungarian IMC, HDSYS Kft. pp. 87-91. GRÓSZ, G. SÁRDI, K. BERKE, J. (2009): Digital leaf area measurement and its application in practice. Georgikon for Agriculture. (közlésre elfogadva) 24