Identification and counting of soybean aphids from digital images using particle separation and shape classification

Aphids population on soybean plants, usually assessed by manual counting, is essential to make pesticide application decisions. Pesticide is applied if the aphid counts exceed the economic threshold of 250 per plant. Manual counting is time-consuming, laborious, and causes visual fatigue. The objective of this study was to develop a method based on computer vision technique to count aphids on soybean leaves. The aphids infested soybean trifoliate were clipped from the greenhouse experiment at three infestation rates (low, medium, and high). Images were captured in the laboratory with three cameras (DSLR, consumer-grade digital camera, and smartphone camera) at two illumination conditions (sunny, and cloudy). The images were processed using a two-stage approach of segmentation followed by classification. In the first stage, image thresholding was performed with marker-controlled watershed segmentation for particle separation to identify the different objects in the image. In the second stage, the identified objects (aphids, exoskeleton, and leaf spots) were classified and counted using shape analysis. The proposed method not only identifies individual aphids, but also has the capability of identifying/resolving touching or overlapped aphids. This approach enables rapid automatic counting (<2 s), after loading the image, compared to manual counting (∼5 min). The system efficiency can be improved through better quality of the image in terms of resolution, contrast, and focus. The accuracy of detecting aphids using image processing technique compared with manual counting gave a good linear fit (R²=0.847).