Objective In the operation of cleaning river floating objects using robots limited computational capacity and complex river environments often lead to the missed detection of small targets. To address these issues an improved YOLOv7 object detection model that balances accuracy and lightweight design is proposed based on YOLOv7. Methods First a small target detection layer with a size of 160×160 was added to the original YOLOv7 to enhance the feature learning ability for small objects. The BiFormer attention mechanism was introduced into the neck of the original network model to improve detection performance for small targets thereby reducing missed detections during river floating object detection. Second the ELAN module in the backbone was replaced with PELAN to reduce the number of parameters and computational overhead thus increasing the detection speed. On this basis the MPDIoU loss function was incorporated to accelerate the convergence and improve the accuracy of bounding box regression and enhance the robustness of the network. Results The improved model achieved a mean average precision mAP of 71. 5% on the FloW subset from Oka-Zhibo which was 4. 7% higher than that of the original YOLOv7 model. Moreover the number of parameters and computational cost were reduced by 11% and 7. 3% respectively. Conclusion Comparative experiments show that the overall performance of the improved model is better than that of the original network model and traditional classic target-detection network models. The proposed model can reduce the model complexity while improving the accuracy making it more suitable for practical applications.