Abstract: A improved model ECSC-YOLO based on YOLOv8m is proposed to address the detection difficulties caused by changes in lighting, target occlusion, and attitude angle in object detection tasks. Firstly, design the EAConv module to enhance feature extraction capability and suppress interference from overexposed or low light areas by embedding an efficient channel attention mechanism. Secondly, the CoordAtt module is improved by introducing a downsampling strategy for the first time, which reduces the size of the feature map while maintaining contextual semantic consistency, significantly enhancing the network’s sensitivity to overlapping targets before and after. Replace the original SPPF module with an improved SPPFCSPC-E module to expand the model receptive field and enhance the multi-scale feature aggregation effect. Finally, to address the issue of precise positioning of the target box, the WIoU loss function is used to optimize the detection head, resulting in more accurate prediction results. The experimental results show that the improved ECSC-YOLO model achieves an average accuracy of 64.8% on the PASCAL VOC2007 dataset, and improves accuracy by 1% ~ 3% compared to existing mainstream algorithms in complex environments, providing an efficient and reliable solution for object detection tasks.