A Markov Random Field Model for Image Segmentation of Rice Planthopper in Rice Fields

It is meaningful to develop the automation segmentation of rice planthopper pests based on imaging technology in precision agriculture. However, rice planthopper images affected by light and complicated backgrounds in open rice fields make the segmentation difficult. This study proposed a segmentation approach of rice planthopper images based on the Markov random field to conduct effective segmentation. First, fractional order differential was introduced into the extraction process of image texture features to gain complete texture information of rice planthopper images. Observation data modeling was established by a combination of image color features and texture features to overcome the disadvantages of insufficient image texture information. Finally, the improved potential function models, the neighborhood relationship between the pixel labels, and the attributes of pixels were defined. The segmentation results were assessed by quantitative evaluation. The experiments showed that the proposed improved approach in the study was more robust, especially with the changes in the illumination condition. This approach can effectively improve segmentation accuracy and promote vision segmentation results of rice planthopper images.