Identification Method of Mature Wheat Varieties Based on Improved DenseNet Model

Wheat is a crucial grain crop in China, yet differentiating different wheat varieties at the mature stage solely through visual observation remains challenging. However, the automatic identification of wheat varieties at the mature stage is very important for field management, planting area, and yield prediction. In order to achieve accurate and efficient recognition of wheat varieties planted in wheat fields, a recognition method based on an enhanced DenseNet network model is proposed in this study. The incorporation of SE and ECA attention mechanisms enhances the feature representation capability, leading to improved model performance and the development of the SECA-L-DenseNet model for wheat variety recognition. The experimental results show that the SECA-L-DenseNet model achieves a classification accuracy of 97.15% on the custom dataset, surpassing the original DenseNet model by 2.13%, which demonstrates a significant improvement. The model enables the accurate identification of wheat varieties in the field and can be integrated into applications for the automated identification of varieties, planting area estimation, and yield prediction in harvester equipment.