<p>Accurate and efficient identification of hazard targets is crucial for enhancing the operational autonomy of rescue robots and ensuring human safety during geological hazard rescue operations. However, existing detection models struggle to simultaneously meet the requirements of high accuracy and lightweight design when addressing the challenges of extreme scale heterogeneity in geological hazard targets, weak discriminative features, and complex background interference. This limits their deployment on resource-constrained edge devices such as rescue robots. To address this issue, this paper proposes a lightweight Depthwise Separable Selective Kernel Attention (DSSKA) module, and based on it, constructs the DSSKA-YOLOv8n detection model. By employing multi-scale depthwise separable convolutions and an adaptive weight fusion mechanism, the proposed module significantly reduces the number of parameters while enhancing the perception and discrimination capabilities for multi-scale targets with weakly discriminative features. Experimental results on a self-constructed geological hazard target detection dataset demonstrate that compared to the baseline YOLOv8n model, our model achieves a 3.1% improvement in precision and a 2.9% increase in mAP50-95. Furthermore, compared to a counterpart model with the standard SKA (Selective Kernel Attention) module directly embedded, our model reduces the number of parameters by 61.2%, effectively achieving a synergistic optimization of accuracy and lightweight design. This study provides a promising lightweight algorithmic solution for the visual perception systems of rescue robots, laying the groundwork for subsequent deployment optimization on robot platforms. It holds positive implications for advancing the intelligent development of geological hazard emergency response.</p>