Abstract: To address the problems that existing deep learning-based belt conveyor coal quantity detection algorithms have a large number of parameters, are difficult to deploy on edge computing devices, and lack quantitative detection capability, a belt conveyor coal quantity detection method based on an improved DeepLabv3+ was proposed. MobileNetV2 was used as the backbone network of DeepLabv3+ for feature extraction, which improved computational speed while maintaining segmentation accuracy as much as possible. Considering the directional characteristics of the coal flow and conveyor belt, as well as the elongated strip-like structure of conveyor belt pixel edges, Strip Atrous Spatial Pyramid Pooling (SASPP) was adopted for enhancement, and the SASPP module was fused with a 1×1 convolution and a residual structure to obtain CA-SASPP, thereby enhancing deep feature extraction. The Convolutional Block Attention Module (CBAM) mechanism was incorporated to achieved weighted emphasis on key information in the feature maps. Experimental results showed that, while the mean segmentation accuracy decreased by only 0.36%, the improved DeepLabv3+ model reduced the number of parameters by 85.58% and increased the inference speed to 113 frames/s, which was 12 frames/s higher than that of the original method, achieving significant lightweight performance while maintaining segmentation accuracy comparable to that of the original model. Based on the semantic segmentation results, quantitative coal quantity detection was achieved by calculating the area ratio between the coal region and the conveyor belt region, which provided a theoretical basis for intelligent speed regulation of multi-stage belt conveyors. The improved DeepLabv3+ model was accelerated using TensorRT and deployed on the Jetson Orin Nano edge computing device. Real-time processing and analysis of coal flow images were achieved, reducing the computational burden on cloud servers and meeting the requirements for real-time performance and accuracy in on-site industrial environments.