Grading of Wheat Leaf Rust Based on Edge Detection of Infrared Thermal Imaging

Wheat rust has a great harm to wheat production in worldwide. The rapid monitoring and classification of wheat rust is the basis for scientific production and management, and it is also the prerequisite to realize the treatment of wheat rust as soon as possible. In view of the shortcomings of conventional image detection algorithms, a fast detection and classification method based on infrared thermal imaging technology was proposed. Wheat samples were planted in a growth chamber at the University of Alberta, Canada. Growth chamber parameters settings were as following: temperature (max 15℃, min 11℃), photoperiod (day 12 h), light intensity (10 000 lx), RH (60%~70%). The spring wheat variety (Peace) was susceptible to rust. The infrared thermal imager brand was FLIR E6, USA. Thermal sensitivity was less than 0.06℃; FOV was less than 45°ohorizontal×34°overtical; IFOV was 5.2×10-3 rad; IR was 160 pixels×120 pixels. The infrared thermal imaging of the whole wheat samples were collected to calculate the average leaf temperature of the healthy plants, the submersible plants and the symptomatic plants, and the temperature changes during the invasion of the fungi were detected. Infrared thermography can be used to detect leaf temperature drop caused by pathogen infection at 6d of pathogen infection incubation period, which was 7 d ahead of the naked eye observation of leaf rust spores. The Prewitt operator (PO), Sobel operator (SO), Canny operator (CO) and Laplacian operator (LO) were used to extract the edges of visible light images. The edge extraction results of PO and SO on the lesion area was not satisfactory for the complex noise processing, and the boundary gray area was seriously ghosting. LO and CO were too lean for the edges, the detection accuracy was reduced, and the background error was too large. Obviously, the direct use of conventional edge detection operators cannot meet the ultimate goal of rapid classification of diseases. Therefore, a fast detection and classification method based on infrared thermal imaging technology was proposed. The experiment was divided into two kinds of extraction methods: single leaf and whole plant. When the whole plant was extracted, the flower pot was removed and only the wheat plant was kept for extraction. From the results of the whole wheat extraction, the area of the whole plant disease can be extracted successfully by the method of area occupation ratio calculation based on the temperature edge. The error of the regional extraction results of the single leaf focus was slightly larger than that of the single leaf focus, but the final calculation results were satisfactory. The region below the temperature threshold was extracted from the infrared thermal image which was preprocessed by histogram equalization and median filtering. The ratio of lesion area to total area of plant thermography was calculated after three steps, including temperature division, low temperature region extraction and threshold segmentation. Finally, the correlation analysis was carried out with the disease index. The correlation coefficient was 0. 975 5, the root mean square error was 9.79%, and the overall recognition rate was 90%. The research result showed that the wheat leaf rust classification method based on the infrared thermal imaging temperature information was feasible. It provided the theoretical and method basis for the early scientific application and the establishment of more accurate disease identification expert system. © 2019, Chinese Society of Agricultural Machinery. All right reserved.