Abstract: Downhole charging is a critical procedure in seismic prospecting that utilizes explosive sources. Strict adherence to operation specifications is crucial to ensuring the safety of personnel and equipment. Currently, manual supervision is widely used for monitoring downhole charging operations. However, this approach suffers from inherent limitations such as discontinuous tracking, poor real-time response, and susceptibility to human error, making it difficult to eliminate potential safety hazards. Moreover, the application of existing deep learning methods to intelligent detection in this scenario is challenged by several factors: small target size, low video resolution, and high similarity of operation details. These give rise to difficulties in small object detection, fine-grained action recognition, and real-time performance. To address these issues, this study proposes a novel intelligent detection method for seismic explosive charging based on the YOLO-ResNet architecture. This method begins with decoding video segments of the charging operation. It uses a YOLOv5-based multi-task detection module to extract information on the spatial locations and working postures of both personnel and detonator boxes. This is followed by the integration of a ResNet101-based fine-grained discrimination module to identify key operation details. Particularly targeting critical safety procedures, such as locking the detonator box and personnel grounding to release static electricity, the detection system introduces an innovative action-duration judgment mechanism to enhance the accuracy and reliability of action recognition. Experimental results demonstrate that the proposed method achieves a detection accuracy of 90% for grounding operations and 86% for detonator box locking, while maintaining an inference speed of 45 frames per second. This performance strikes a balance between high accuracy and real-time detection capability. This method not only shows significant application potential in seismic explosive charging operations but also offers a new framework for the intelligent monitoring of complex industrial operations through video analysis.