Addressing the current issues of limited research and low detection accuracy in snake detection, this paper proposes an improved YOLOv5s detection algorithm - YODE-FEIM. Firstly, based on YOLOv5s, the FasterNet Block is combined with the C3 module in the backbone and neck, and the EMA attention mechanism is added at the end of the module, enhancing the feature extraction capability in snake images while reducing parameter computation. Secondly, the detection head is replaced with the RT-DETR detection head, accelerating model convergence and improving detection accuracy. Finally, the Inner-MPDIou loss function is introduced, utilizing boundary regression for localization to enhance the model's detection precision. Experimental results show that on the self-made ChineseSnake dataset, the proposed YODE-FEIM model achieves a precision (P) of 92.7% and a mean average precision (mAP) of 90%, demonstrating high accuracy and providing support for snake detection in wild environments.