A novel algorithm for the automatic segmentation of rice planthoppers in rice fields using image processing

摘要

The diagnosis of rice planthopper pests based on imaging technology is an efficient means to develop intelligent agriculture. Automatic contour extraction is an important pretreatment technology for the early identification and classification of rice planthopper pests. However, the segmentation of rice planthoppers is made difficult by the heterogeneous texture of sensed images resulting from the curtain serving as the background. Thus, this study developed a novel method based on digital image processing to segment rice planthoppers with a curtain under paddy field conditions. First, the output phase of the transform was used for edge detection on the basis of a nonlinear dispersive phase operation applied to the image. Second, to further increase the robustness of the proposed method, we calculated the entropy of the edge directions in a small neighborhood window to convert the gray-level image into an entropy image because the background edges showed consistent edge directions and a pest's region presented a high variation of edge directions. Then, we proposed a mean shift method that is based on color features, entropy features, and edge features, which were integrated to enhance the target information and thereby reduce the localization error of object segmentation. Experimental results showed the successful implementation of the proposed method in the segmentation of rice planthoppers. Moreover, the integration of the aforementioned features improved the robustness of the proposed algorithm to changes in illumination conditions.