Abstract: In blasting operations within rock tunnel excavation faces, the current practice of manual explosive charging or robotic arm operation by experienced personnel struggles to ensure both operational effectiveness and safety. The intelligent advancement of explosive-charging robotic arms is critical to achieving safe and efficient explosive filling in tunnel blasting operations, with the blasthole image detection algorithm serving as the core component of the intelligent control system for these robotic arms. To achieve intelligent control of the explosive-charging robotic arm, ensure the accuracy of blasthole image detection, and reduce the power consumption of the control device, enabling the embedded control device of the robotic arm to meet the safety requirements of intrinsically safe electrical products, a lightweight blasthole detection and localization algorithm Mv3-SCD, is proposed. In terms of blasthole detection accuracy, the algorithm addresses the issue of false detection caused by the influence of surrounding rock backgrounds and rock shadows, as well as the problem of missed detection due to limited contextual information and identifiable features of blastholes in images. Firstly, a blasthole detection head structure is designed, which utilizes a high-resolution detection head to reduce the loss of blasthole features caused by excessive downsampling. Secondly, the Mv3_Block is introduced to enable the algorithm to possess strong semantic abstraction capabilities for blastholes even at shallow feature levels. By incorporating an atrous spatial pyramid pooling module, the receptive field is expanded to capture fine-grained differences between blastholes and shadows formed by rock occlusion in complex surrounding rock backgrounds. Finally, the loss function is optimized to improve the accuracy of blasthole bounding box regression. To tackle the issues of large parameter size and low frames per second (FPS) in the blasthole image detection and localization algorithm network model, a lightweight Sc_C2f module is proposed to optimize the network structure. To validate the effectiveness of the algorithm, both subjective and objective analyses of Mv3-SCD are conducted. Compared with the minimum baseline model, the Mv3-SCDn blasthole algorithm has the best blasthole detection effect, the number of blasthole detection model parameters is reduced by 7.17%, and the detection speed is increased by 45.44%. Experimental results indicate that the proposed algorithm effectively enhances the precision of intelligent explosive-charging robotic arms and achieves a higher level of network model lightweighting.