A Study on a Wheat Image Recognition System Based on Transformer and Mamba Networks with Light Enhancement

In this paper, the application of machine vision and deep learning technologies in smart agriculture, particularly in crop disease and pest identification, crop growth monitoring, and crop lodging detection is explored. By utilizing visible light images of wheat canopies obtained from drones, a deep learning model tailored for low-light conditions has been constructed. This model integrates Transformer networks and Mamba networks, gradually improving image quality and recognition accuracy through multi-layer processing. The experimental results show that this model achieves outstanding performance in wheat recognition tasks, with the precision (precision) reaching 0.831, the mean intersection over union (Miou) reaching 0.761, the recall reaching 0.751, and F1 score reaching 0.821, significantly outperforming other classical models. These results demonstrate the effectiveness and superiority of the model in practical applications.