Semantic segmentation-based method for rice leaf blast segmentation and grading
A method based on semantic segmentation for rice leaf blast segmentation and grading was proposed to address the issues of efficiency and accuracy in traditional methods.Firstly,images of rice leaves from the CO39 variety were collected,and the leaf and lesions were annotated using the Labelme annotation software to create a leaf dataset.Then,three rice leaf segmentation models,namely VGG16-UNet,ResNet50-UNet,and MobileNetV2-DeepLabV3+,were constructed by using different convolutional neural networks as the backbone feature extraction networks.These models were used to segment rice leaves and lesions.Based on the grading criteria and grading formula of rice blast,the disease grade of rice leaves was determined.During this process,the segmentation performance of the three models was compared.The results showed that the VGG16-UNet model performs the best,achieving an average pixel accuracy,average intersection over union,and F1 score of 86.87%,80.68%and 88.48%,respectively.It effectively met the practical requirements of rice leaf blast segmentation and grading.The proposed method provided a theoretical basis for developing an intelligent grading system for rice blast and serves as a referenced for grading research on other crop diseases.
NETL
NSTL
万方数据
