Dual-path model for irrigation facility detection via UAV remote sensing

Accurate and efficient condition detection of irrigation facilities is crucial for ensuring stable crop production and enhancing food security. However, traditional detection methods often fail to meet the demands for refined monitoring in large-scale irrigation areas. To address this issue, this study proposed an irrigation facility condition detection system based on UAV low-altitude remote sensing and the DP-ResNet50 model to identify four common states of irrigation canal systems. First, with ResNet-50 as the backbone network, the CalibratedFocal Loss function was adopted to mitigate biases arising from imbalanced class distributions and reduce overconfident predictions in ambiguous scenarios. Second, the Dual-Path and Fusion Collaborative Module (DP-FCM) was designed to enhance the ability to capture differentiated local features, such as silt and gravel, through the complementary combination of a general backbone and a lightweight discriminative path. Finally, to further address the high-dimensional redundancy and low-dimensional confusion in irrigation canal state features, the Enhancement and Classifier Integrated Module (EC-IM) was proposed to realize more effective feature mapping and state discrimination. Experimental results demonstrate that the DP-ResNet50 model achieved favorable performance in irrigation canal condition detection, with an accuracy of 0.9413, alongside macro precision, macro recall, and macro F1-score of 0.9250, 0.9195, and 0.9211, respectively. Furthermore, compared to other classic models, the DP-ResNet50 model demonstrates higher and more balanced recall rates for individual classes. The results of this study can provide a promising solution for the intelligent monitoring of smart agricultural irrigation systems.