Plant recognition method based on deep residual network and attention mechanism
Download as PDF
Xiaomeng Lu, Ming Zhang
This paper introduces an improved version of ResNet18network recognition model, which aims to solve the shortcomings of traditional plant recognition methods, such as the complicated steps of manual feature extraction, long time and low accuracy. The model uses the attention mechanism and transfer learning technology, and combines the SE-Net module to improve the model. Specifically, SE-Net modules are added to each residual block to enhance the weight of useful features and reduce the effect of useless features such as noise, thereby improving feature extraction capabilities and enhancing the robustness of the model. In addition, the model uses the trained parameters on the ImageNet data set to apply to the expanded plant image data set to improve the generalization ability of the model. In order to further reduce the influence of overfitting, the network structure is adjusted, and the batch standardization layer and activation function layer are placed in front of the convolution layer to enhance the regularization effect of the model. The experimental results show that the model performs well in recognition accuracy and has certain guiding significance.
plant image recognition, residual network, Feature extraction, Attentional mechanism, Transfer learning