Precision Agriculture
Seyedeh Arefeh Hosseini; Hassan Masoudi; Seyed Majid Sajjadiyeh; Saman Abdanan Mehdizadeh
Abstract
Introduction Nitrogen is one of the essential elements for plants and is consumed more than other elements in plant nutrition. Nitrogen is an important component of the chlorophyll molecule and is present in the chlorophyll structure as a protein. Without nitrogen, plant growth decreases significantly. ...
Read More
Introduction Nitrogen is one of the essential elements for plants and is consumed more than other elements in plant nutrition. Nitrogen is an important component of the chlorophyll molecule and is present in the chlorophyll structure as a protein. Without nitrogen, plant growth decreases significantly. This research was carried out to estimate the amount of nitrogen and chlorophyll of sugarcane leaves from color indices extracted from digital aerial images taken by a quad-copter at two 5 and 10 m altitudes in the fields of Debal Khozaie sugarcane agro-industry company, Khuzestan, Iran. The images used for this research are from three farms with different growth stages. Materials and Methods The imaging was carried out using a quad-copter, the Phantom 3 professional model, at two heights (5 and 10 meters) from the specified points in the fields. After taking the photos from all marked points by the quad-copter camera, four healthy cane branch - with 45 cm distance from each other - were picked at each point and placed in plastic bags. Then, samples were immediately transferred to the laboratory to measure the leaf chlorophyll value, moisture content and the amount of nitrogen. Using a hand-held chlorophyll meter (SPAD-502 model), the leaf chlorophyll index was measured and recorded at each point. After drying the samples, the nitrogen levels were measured using the manual Kjeldahl method. The designed image processing algorithm, to extract color indices from sugarcane fields' images, had these steps: image transfer, preprocessing, image smoothing, noise, and background removal, extracting and selecting of image attributes. After using the image processing algorithm, the color indices of the fields' images were obtained; then the relationship between color indices and nitrogen and chlorophyll content of sugarcane leaves were determined using multivariate regression. The preparation of the data was done in Excel 2013 software and the development of multiple regression equations in SPSS v.21 software. The student t-test was used to compare the performance of regression models in the prediction of nitrogen and chlorophyll content with real values. Results and Discussion Based on the results of the measurements, the dispersion of nitrogen was not uniform throughout and between the fields. The least nitrogen dispersion was in the first growth period and the greatest one in the second growth period. None of the fields had uniform dispersion in the chlorophyll content. The least dispersion was observed in the second growth period and the highest dispersion in the third growth period. Based on the Pearson correlation statistical analysis - from 48 features extracted by image processing including mean, variance, skewness, and peak value of each image color indices in RGB, HSV, HIS, and Lab color spaces - only 24 features were selected to determine the regressions equations. These indices had a correlation with the amount of nitrogen in sugarcane leaves. In the images of 5 meters height, the obtained regression equation for nitrogen estimation was significant at 1% probability level and had a 74.3% determination factor. The determination factor of the five regression equations presented for the images taken from 10 m height were 71, 74, 77, 79, and 82 percent. Also, all the regression equations were significant at 1% probability level, so these relationships are valid and can be used to estimate the amount of nitrogen in sugarcane from 10 m height. By increasing the number of color indices, the accuracy of the regression model in the estimation of nitrogen levels was increased. Accuracy of the 10 m regression model for estimating the amount of nitrogen in sugarcane was higher than the 5 m regression model. All four regression models presented for estimating the chlorophyll of leaf based on color indices of images taken from 5m height were significant at 1% probability level. The obtained determination coefficients for these models were 26, 45, 55, and 62%. By increasing the number of color indices, the accuracy of the regression model was increased for the estimation of chlorophyll content of the leaf. Also, the presented regression model for the estimation of leaf chlorophyll based on color indices obtained from 10 m height images was significant at 1% probability level. The determination factor for this model was 69%, which is more than the determination factor of the most accurate regression model presented for 5m height images. The regression model presented for estimating the sugarcane nitrogen content from leaf chlorophyll was significant at 1% probability level. The amount of determination factor for this model was 68%, which is very close to the amount reported by the Debal Khozaie sugarcane agro-industry company, Khuzestan, Iran. Conclusion Thecomparison of the results of regression equations with real values showed that nitrogen prediction with regression model for 5 m height images and two regression models for 10 m height images had no significant difference with each other. Also, the results of sugarcane nitrogen estimation using the leaves chlorophyll was not significantly different from the actual nitrogen content of leaves. On the other hand, chlorophyll prediction was performed by two regression models for 5 m height images and the regression model for 10 m height images were not significantly different from the actual amount of leaves chlorophyll. Therefore, the presented regression equations are valid and reliable and using these relationships can help know the state of nitrogen and chlorophyll in sugarcane fields.
