Image Recognition of Crop Diseases and Pests Based on Deep Learning
To address the problem of low accuracy and efficiency in the image recognition of crop disea-ses and pests,a multi-granularity feature segmentation model(MGFSM)based on convolutional neural networks and transfer learning is proposed.This model utilizes ResNet-50 as the backbone network and le-verages pre-trained weights for parameter sharing to enhance recognition accuracy and stability.MGFSM comprises a global branch that learns the appearance model of the entire image and a local branch that ad-dresses complex backgrounds and occlusion issues.Additionally,it jointly utilizes metric loss and classifi-cation loss functions to overcome the challenges of large intra-class distances and small inter-class dis-tances,thereby improving the discriminability of disease and pest images.The model's adaptability to dif-ferent types of diseases and pests is enhanced by transferring knowledge to the augmented target data through transfer learning.Experimental results demonstrate that the MGFSM model exhibits high recog-nition accuracy and generalization capabilities on the PlantVillage dataset,making it suitable for practical applications on mobile and embedded devices.
万方数据
维普
