AI Detection Method for River Water Level Based on Image Segmentation Processing

River water level detection is accomplished by sensors. However, the bending of the river channel, width changes, riverbed morphology and other factors result in errors in the water level images monitored by the sensors, which in turn affects the accuracy of small-scale feature mapping. For this reason, this paper proposes a river water level AI detection method based on image segmentation processing. The river water level image is preprocessed by mean filtering and histogram equalization to remove noise and enhance contrast. The Retinex defogging algorithm is utilized to further improve the image quality. An image segmentation model based on L1-[Formula: see text]L2 regularity is constructed, and the image segmentation is realized by solving DCA and ADMM algorithms. In the traditional YOLOv5 network structure, the meso-scale and large-scale feature mappings are fused into the small-scale feature mapping by weighted summation to enhance the spatial information, detail information and semantic information transfer capability of the small-scale feature mapping, and the water level detection is carried out based on the improved YOLOv5. The experimental results show that when the method in this paper deals with different backgrounds, light and reflection images, the segmented accurate distinction between the river water level region and other background regions has obvious boundaries. The accuracy of water level detection is always around 98%, and the recall rate is always above 96%, which is practical.