Foreign Object Detection On Coal Flow Surface Based On Semi-supervised Generative Adversarial Learning

During the transportation of coal from mining to the ground,it will be mixed with foreign object such as bolts and large coal stones,which will easily cause damage to the transmission device,lead to coal mine safety production accidents,and seriously threaten the personal safety of workers.The detection of foreign object in the coal flow of the belt conveyor is of great significance to realize the safe and efficient production of coal mines and promote the construction of intelligent mines.Traditional object detection methods based on supervised learning require a large amount of labeled image data,but coal flow images containing foreign objects in real industrial and mining scenarios are difficult to capture and the types of foreign object are varied,which limits the effectiveness and applicability of the method.In view of this,this thesis uses a semisupervised learning strategy to study a coal flow surface foreign object detection model based on generative adversarial learning,which mainly includes the following contents:Aiming at the problem of redundant data and unbalanced samples of normal coal flow,the optimization of data-cleaning,feature extraction methods and algorithm flow is studied.First of all,considering the normal sample redundancy phenomenon caused by a large number of no-loads and stalls in the data,the perception Hash(perception Hashing,p Hash)algorithm is used to clean the modeling data,and the frequency components of the image are analyzed and screened by discrete cosine transform After weight loss.Then,the foreign object recognition based on the combination of autoencoder model and clustering is studied.During the training process of autoencoder model,the model is forced to pay more attention to the feature information of the image itself,which can effectively alleviate the impact of sample imbalance on feature extraction.Finally,the foreign object data is detected on the subclass of the clustering result,and the foreign object recall rate reaches 59% on the test set.In order to further improve the model’s ability to judge unknown foreign matter categories,a coal flow surface foreign matter detection algorithm based on semisupervised generative adversarial learning is proposed,which completely gets rid of the dependence on abnormal coal flow images in the training stage to improve the robustness of unknown foreign matter category detection.On the one hand,considering the information loss of the generator module in the process of encoding to decoding,a simplified cross-channel attention mechanism is proposed to guide the network to weaken useless background information while increasing the parameter weight of the coal flow movement area.On the other hand,a multi-scale feature fusion module is proposed,which uses multi-scale feature maps to extract spatial and channel features under different receptive fields.The reasoning time of a single image is only 0.0071 s,which meets the needs of actual coal mine detection.In addition,in order to verify the effectiveness and applicability of the proposed method,experiments are carried out on self-built datasets and two public datasets.The results show that the method can effectively use different levels of feature information to obtain high-quality recognition results,which is superior to the latest foreign object detection model based on generative adversarial networks.The above algorithms are integrated and packaged,and a set of coal flow foreign object detection software with convenient operation and friendly interface is developed based on Python and Py Qt.It mainly includes modules such as real-time detection,historical data detection and historical detection data query,which can realize model calling,video streaming and Functions such as image detection and data storage are helpful for the application of foreign object detection technology in coal mining enterprises.This thesis has 30 figures,13 tables and 91 references...