Research on Foreign Object Detection Algorithm for Coal Conveyor Belt Based on Deep Learning

To accelerate the modernization and intelligence of the coal industry and replace the existing detection methods that rely on traditional physical characteristics such as rays and spectra, this paper proposes an improved target recognition algorithm based on YOLOv8n to enhance the safety and stability of underground coal mine production. The algorithm presented here is an improved iteration of YOLOv8n: First and foremost, the study introduces the cutting-edge SPD-Conv spatial encoding method into the backbone architecture, enabling more effective capture of multi-scale features, this approach enhances deep learning model performance via more effective image data processing. It converts the spatial information of the image into depth information, enabling the convolutional neural network (CNN) to learn image features more effectively. Secondly, in the neck network, a high-level feature selection and fusion pyramid structure is adopted to enhance the multi-level feature fusion capability. The high-level features are used as weights to filter the low-level feature information through the channel attention module. The channel attention module integrated into the algorithm employs high-level features as weights to selectively suppress irrelevant low-level features and enhance target-related responses, thereby enhancing semantic consistency in multi-scale feature fusion. and then combines the results with high-level features to enhance the feature expression capability of the model. Experiments show that this algorithm performs better with the same sample size, with a precision (P) of 82.4% and an average precision ([email protected]) of 79.3%, meeting the requirements of accuracy and real-time performance for target detection tasks on coal mine conveyor belts.