Rice disease identification system in the field environment based on Cutout-ResNet50
Aimed at the problems of rice disease images in the field environment,such as uneven illumination,obvious changes in brightness,missing target features and overlapping noise due to occlusion,and few rice data sets in the field environment with poor quality,an improved ResNet50 algorithm is proposed for rice disease identification in the field environment and the recognition sys-tem is designed.Based on the traditional ResNet50 algorithm,the transfer learning technique is adopted to transfer the learning knowledge across domains to alleviate the overfitting phenomenon caused by insufficient and unbalanced data sets.The Cutout enhancement method is utilized to filter the feature information immediately to simulate the complex field environment and enhance the gen-eralization ability of the algorithm.The cosine annealing optimization strategy is adopted for the learning rate to improve the stability of the algorithm.The results show that the improved Res-Net50 algorithm has a recognition accuracy of 97.24％on a small rice disease dataset,which is sig-nificantly higher than that of the traditional ResNet50 algorithm,and the improved method also has an enhancement effect on other convolutional neural network algorithms such as VGG16,GoogLeNet and MobileNetV3-large.The model is deployed in the system,which can provide tech-nical reference for the development of rice disease identification in practical application engineering.
NETL
NSTL
万方数据
