Developing an image processing based algorithm to detect and count soybean aphids under field conditions

Abstract. Soybean aphid (Aphis glycines) is one of the most important pests in soybeans in the north central region of US. Aphids have the potential to multiply rapidly. High infestations can stunt plant growth and development causing serious yield damage up to 95%. The current management practice of regular scouting and counting aphids on a plant is a labor-intensive and time consuming process. A count of 250 aphids on a single soybean plant is considered as an economic threshold for potential yield damage, and therefore insecticide application. Image processing techniques have been used in the past for automated detection and counting of insect pests. In this study, a previously developed image processing algorithm to detect and count soybean aphids in the controlled environment in a greenhouse facility was modified for data collected from a soybean field. The study was conducted in a soybean research fields at the North Dakota State University during the 2015 growing season. Images of the top and bottom sides of infested soybean leaves were captured with different cameras under daylight. The cameras used in the study included a 7.2 MP SONY^TM regular digital camera, an Apple iphone^TM 6 with and without a macro lens and a Nokia^TM Lumia 1020 cell phone. The captured images were processed in MATLAB^TM software to identify and count aphids. Three different segmentation methods were used to discriminate aphids from other objects on the leaf based on hue, object area and aspect ratio of aphids. The aphids counted with the sensing systems were compared to the count generated manually by a trained expert to evaluate the accuracy of the algorithm. The counts generated by the digital imaging method correlated very well with the manual counts. The study also showed that images captured with an inexpensive digital camera and different cell phone cameras were satisfactory (r>0.91). Efforts are still under progress to improve the algorithm for images captured under cloudy condition in the fields.