Intelligent recognition algorithm for rice diseases in farmland based on multileaf segmentation

Abstract

Timely and accurate identification of specific types and development trends of rice ( Oryza sativa L.) diseases in farmland is a critical prerequisite for implementing scientific control measures and reducing yield losses. Aiming at the problem of low recognition accuracy for rice diseases in farmland under natural lighting conditions, this paper proposes an intelligent recognition algorithm for rice diseases in farmland based on multi‐leaf segmentation. First, we propose the multi‐leaf segmentation model based on edge aware guidance (EAG‐Mask2former) for rice canopy images in field, which can segment at least three to five complete and independent rice leaf images. Subsequently, we propose the rice disease intelligent recognition algorithm based on ConvNeXt V2 network for rice leaf images, which can quickly and accurately identify the rice disease categories, thereby achieving rice disease recognition in farmland. Experimental results demonstrate that the proposed algorithm achieves both high multi‐leaf segmentation accuracy and model lightweightness, with mean intersection over union of 81.70%, and only 44.0 M parameters. Besides, the disease recognition accuracy reaches over 97.17% for both public images and output images of the field canopy images, with an inference time of 7.91 ms/img, achieving rapid recognition for rice diseases in farmland. In the actual field scenarios, the overall accuracy of the end‐to‐end disease recognition algorithm reaches 87.81%, demonstrating its high precision and practicality. Further experimental results for corn ( Zea mays L.) diseases reveal that the accuracy of this algorithm for identifying corn diseases is as high as 98.26%, presenting a feasible technical solution for intelligent recognition for crop diseases in farmland.