Segmentation and Grading Method of Potato Late-Blight on field by Improved Mask R-CNN

Potato late-blight is a severe plant disease caused by the fungus Phytophthora infestans. This disease can quickly spread in a short period, leading to a significant decrease in potato production and posing great harm to the global potato planting industry. Therefore, an effective method for identifying late epidemic diseases has important practical significance. This paper proposed potato late-blight segmentation and grading method for the image with multiple leaves collected on farm. The improved Mask R-CNN network is presented to segment potato leaves at the pixel level and then grade late blight based on the segmentation information. Then, the segmentation effect of the model on the image will determine the accuracy of late blight evaluation. The neural network achieved a Mask recognition rate of 88.99% and 91.35% for leaves and lesions. It is possible to achieve disease grading operations for late epidemic diseases.