煤矿井下带式输送机的煤流中掺杂着锚杆、槽钢、木条、矸石和大块煤等异物，不仅会造成皮带划伤，严重时发生撕裂或断带，甚至会出现严重的安全事故。随着国家煤矿智能化和人工智能技术发展，煤矿井下开始采用机器视觉技术对煤矿带式输送机异常状态进行识别。因此，本文采用机器视觉和深度学习技术对煤矿井下带式输送机异物智能识别方法进行研究，对煤矿带式输送机安全、高效、可靠运行和提高煤矿生产效率具有重要意义。

针对煤矿井下带式输送机异物导致输送带损伤问题，提出一种融合图像清晰化的改进YOLOv5煤矿井下带式输送机异物智能识别方法，将该方法生成的模型进行优化后部署在煤矿井下NVIDIA Jetson AGX Orin上，并开发了煤矿带式输送机异物智能识别上位软件，能够实现煤矿井下带式输送机异物高效、智能识别。

针对煤矿井下照明设施安装不均匀以及带式输送机高速运行造成监控视频图像出现明暗不均、运动模糊问题，使用基于二维伽马函数的图像自适应增强方法来提高图像暗处的亮度，抑制亮度过高部分，并提出改进DeblurGANv2的去运动模糊算法对视频图像进行快速去运动模糊处理，与多种目前广泛使用的图像增强和去运动模糊算法相比，本文方法最优，为后续煤矿井下带式输送机上异物的智能识别提供了清晰的视频图像。

针对煤矿井下带式输送机异物识别的漏检和误检问题，提出了一种融合图像清晰化的改进YOLOv5带式输送机异物智能识别方法，采用Channel Attention Module自动为每个异物特征通道授予不同的权重系数，从而强化异物重要特征，再通过Spatial Attention Module为每个空间位置进行加权输出，弱化异物杂乱背景干扰，进一步提高了异物特征表达能力，使异物识别模型对异物产生注意力偏重。Adaptively Spatial Feature Fusion可以对异物图像高层的语义信息进行更好的利用，对底层异物轮廓和异物形状等信息进行更加充分的融合，使得模型中包含了更丰富的异物特征信息，从而实现了对煤矿井下带式输送机异物的精确识别。改进后异物识别模型的识别精确率为96.9%，检测速度为25.64FPS。此外，针对煤矿井下带式输送机监控视频在地面进行异物识别的延迟问题，本文通过TensorRT引擎对异物智能识别模型进行优化，并将其部署在井下的NVIDIA Jetson AGX Orin设备上，实现了在煤矿井下对带式输送机异物的实时识别。

为了便于集控中心对煤矿井下带式输送机异物智能识别情况进行监测，开发了基于PyQt的带式输送机异物智能识别系统软件，通过实验验证了系统软件的监控相机连接功能、视频图像清晰化功、异物智能识别功能和异物识别结果存储与查询功能的有效性。

运用黄陵煤矿、韩家湾煤矿井下现场带式输送机视频对煤矿井下带式输送机异物智能识别系统进行了实验验证，结果表明该系统改善了煤矿井下带式输送机异物识别时的漏检和误检问题，并且带式输送机异物智能识别系统在实验室带式输送机的实验中识别精确率为96.39%，识别速度为25.64FPS，实现了带式输送机异物精确、实时的识别。

The coal flow of underground belt conveyors in coal mines is mixed with foreign objects such as anchor rods, channel steels, wooden strips, gangues, and large pieces of coal, which not only cause belt scratches, but also serious tearing or breakage, and even serious safety accidents. With development of national coal mine intelligence and artificial intelligence technology, machine vision technology has been used underground to identify abnormal states of coal mine belt conveyors. Therefore, machine vision and deep learning technology are adopted to study intelligent recognition method of foreign objects on coal mine underground belt conveyors, which is of great significance for safe, efficient, reliable operation of coal mine belt conveyors and improvement of coal mine production efficiency.

In response to problem of foreign objects causing damage to conveyor belt on coal mine underground belt conveyors, an improved YOLOv5 intelligent recognition method with image clarity processing for foreign objects on belt conveyors in coal mines is proposed. The model generated is optimized and deployed on NVIDIA Jetson AGX Orin underground coal mine, and an intelligent recognition software for foreign objects on coal mine belt conveyors has been developed, which can achieve efficient and intelligent recognition of foreign objects in coal mine underground belt conveyors.

In response to problem that uneven brightness and darkness and blurry of monitoring video caused by uneven installation of lighting facilities in coal mines and high-speed operation of belt conveyors, an image adaptive enhancement method based on two-dimensional gamma function is used to improve brightness of dark areas in images, suppress excessive brightness, and use the improved DeblurGANv2 motion blur removal algorithm to quickly remove motion blur of videos. Compared with various widely used image enhancement and motion blur removal algorithms, the method proposed has the best results, providing clear video images for intelligent recognition of foreign objects on underground belt conveyors in coal mines.

An improved YOLOv5 belt conveyor foreign object intelligent recognition method with image clarity is proposed to address issues of missed and false detections in foreign objects recognition of underground belt conveyors in coal mines. The Channel Attention Module is used to automatically assign different weight coefficients to each foreign object feature channel, thereby enhancing important features of foreign objects. The Spatial Attention Module is then used to weight output for each spatial position, weakening interference of cluttered background of foreign objects further improves feature expression ability of foreign objects, causing foreign objects recognition model to pay more attention to foreign objects. Adaptive Spatial Feature Fusion can make better use of the Semantic information at the top level of foreign object images, and more fully integrate information such as contour and shape of foreign object at the bottom level, so that model contains more abundant foreign object feature information, and finally realizes efficient and precise recognition of foreign objects on belt conveyor underground in coal mine. The recognition precision of the improved foreign objects recognition model is 96.9%, and detection speed is 25.64 FPS. In addition, in response to delay problem of foreign object recognition on ground in monitoring videos of belt conveyors in coal mines, the intelligent foreign object recognition model is optimized with TensorRT engine and deploys it on the NVIDIA Jetson AGX Orin equipment underground, achieving real-time recognition of foreign objects on belt conveyors in coal mines.

In order to facilitate centralized control center to monitor intelligent recognition of foreign objects on coal mine belt conveyors, an intelligent recognition system software  baed on PyQt for foreign objects on belt conveyors has been developed. The effectiveness of system software's monitoring camera connection function, video image clarity function, foreign object intelligent recognition function, and foreign object recognition result storage and query function has been verified through experiments.

The intelligent recognition system for foreign objects on underground belt conveyors in coal mines is experimentally validated using on-site belt conveyor videos from Huangling Coal Mine and Hanjiawan Coal Mine. The results presented that system address problems of missed and false detections of foreign object recognition on underground belt conveyors in coal mines. The recognition precision of intelligent recognition system for foreign objects on belt conveyors is 96.39% and recognition speed is 25.64 FPS in laboratory experiments, achieving precise and real-time recognition of foreign objects on belt conveyors.