Research On Forest Phenology Recognition Based On Attention Mechanism

In view of the insufficient feature extraction of current phenological phase recognition methods and the failure to distinguish key features,resulting in low recognition accuracy and poor generalization performance of the method,this paper establishes a phenological phase recognition model based on the attention mechanism.The research designs variety of different realization form of the attention mechanism of structure,the fusion form of channel attention and spatial attention is studied,and a fusion attention module without a fully connected layer is proposed.At the same time,the lightweight attention network model is studied from the perspective of separating convolution and model pruning,which lays the foundation for the expansion of the application scenarios of the phenological period recognition model.In order to verify the effectiveness of the model,the article trains and tunes the model based on the Pheno Cam phenology image data set,takes oak and maple as the research objects,and validates the model effect through experiments.The experimental results show that the introduction of the attention module has steadily improved the generalization of the model,enhanced the model’s ability to recognize finegrained features,thereby enhancing the perception and recognition of subtle differences in forest phenology,improving the model’s ease of use,increasing the recognition accuracy of the confused phenological period.Among the attention module structures of different structures,the CBAM-ECA performance of the attention module without a fully connected layer is the best.The Res Next50-CBAMECA model achieves the highest recognition accuracy and better robustness,and the model passes the residual neural network fully extracts the features of each phenological stage image.The introduction of the attention mechanism improves the efficiency of feature utilization and improves the generalization performance of the model.The migration and application to the 2019 data set performed well,and the accuracy rate in the maple forest research area was 90.87%,the accuracy rate of the oak forest study area is 91.21%,which is better than other model combinations.In terms of research on lightweight networks,experiments have shown that Mobile Net-v2,which introduces the attention mechanism,maintains a high accuracy rate while maintaining low parameters,which are 87.39% and 88.89% in the two research areas.The experimental results demonstrate the effectiveness of the attention mechanism in the field of forest tree phenological phase recognition.The phenological recognition model based on the attention mechanism proposed in this paper solves the problem of poor generalization ability of traditional methods.It has the advantages of high accuracy and strong robustness.It can accurately distinguish the phenological stages of forest trees with small morphological differences.The compressed model can be deployed in scenarios such as embedded devices,and is suitable for long-term observation of forest phenology,thereby providing technical support for precision forestry.