Coal and gangue recognition technology and implementation based on deep learning

Aiming at the problems of low speed, low efficiency, fluctuating accuracy and difficulty in practical application of traditional coal gangue identification methods based on image processing, a coal gangue identification method based on an improved lightweight deep recognition network model is proposed, which is based on the MobileNetV3-large module structure. As a basis, under the premise of ensuring that the volume and complexity of the model parameters are less increased, the performance of the network model is further improved to make it more suitable for the actual production environment of coal and gangue mining or picking. Firstly, CBAM attention mechanism module is used in the model, which has higher characterization ability than SE module used in the original model, and can better improve the feature extraction of coal and gangue images with complex pixel information.Then, the training data set is processed by corresponding complex image enhancement techniques such as color, position and fuzzy image enhancement techniques. On the one hand, it increases the generalization ability of the recognition model to recognize the complex production environment of coal gangue, and reduces the risk of network overfitting. On the other hand, it completes the further expansion of the data set, and finally obtains an improved lightweight deep recognition network model through the above methods. Finally, the improved model is applied to the research and implementation of coal gangue identification technology.The experimental research results show that the coal and gangue identification method based on the improved lightweight deep recognition network model has the advantages of simple structure, easy network training, easy embedding and high recognition accuracy. In the test of coal and gangue identification, compared with the original model, the accuracy is increased by 2.3%, reaching 97.7%, and the recall rate has increased by 2%, reaching 97.8%. The automation degree and production efficiency of the working face and the realization of intelligent coal mining and picking have important application value.