Abstract: In recent years, coal mine robots have become a research hotspot in the field of modern coal machine equipment, and the main coal flow transportation system of most coal mines has basically realized continuity, mechanization and automation, which also puts forward higher requirements for safety monitoring and inspection efficiency in the main transportation roadway, and accurate target detection is a necessary guarantee for intelligent safety monitoring in coal mines, but the existing object detection algorithm is applied to complex and harsh coal mine underground roadway environment, and there is a problem of low target detection accuracy. Aiming at the special working condition detection requirements of low lighting and chaotic environment in the downhole, the target data set in the underground roadway environment was produced, and the dataset annotation was completed and multi-dimensional analysis was carried out. A PT target detection algorithm based on the fusion of pyramid structure and attention mechanism is proposed, and the attention mechanism module is used to replace the convolution module in the pyramid structure, which improves the extraction ability of global features while controlling the amount of feature calculation, realizes the extraction effect of the fusion of local features and global features of the target, and improves the expression ability of the features of the target area of interest in the image. Finally, for the application scenario of underground inspection robot in coal mine, the proposed PT algorithm is compared with the traditional Faster R-CNN and YOLOv4 algorithms. Compared with the mainstream Faster R-CNN and YOLOv4 target detection networks, the PT algorithm has better comprehensive recognition capabilities, and the accuracy of identifying coal mine personnel is increased by 2.90% and 4.30%, the accuracy of identifying underground obstacles is increased by 0.20% and 4.80%, and the accuracy of identifying mine cracks is increased by 4.40% and 8.60%, respectively. The accuracy rate of identifying downhole equipment was improved by 3.00% and 8.70%, respectively. Therefore, the PT target detection algorithm can better adapt to the underground environment, and the target detection algorithm can obtain higher accuracy and detection speed than other algorithms, which can provide theoretical basis and technical support for the construction of underground roadway security control system.