基于Mask R-CNN的近海漂浮垃圾智能识别与清理路径规划系统

Intelligent Recognition and Cleanup Path Planning System for Offshore Floating Debris Based on Mask R-CNN

摘要: 针对近海漂浮垃圾人工清理效率低、成本高和风险大的问题，本文设计了一套基于Mask R-CNN的智能检测与清理路径规划系统。系统采用Roboflow海洋垃圾数据集，包含11类目标、10,000张图像和56,272个标注实例；基于ResNet-50-FPN的Mask R-CNN实现目标检测与实例区域分割，并将清理路径建模为非闭合旅行商问题，采用最近邻贪心算法与2-opt局部搜索进行优化。实验结果表明，模型在验证集上的边界框AP为53.50%，掩码AP为 52.05%；路径规划算法在200个目标规模下耗时约181.8 ms，路径长度缩短约10.5%。同时，本文开发了Web与桌面双模态可视化系统，实现了图像输入、目标检测、路径规划和结果展示的原型化闭环流程，为近海漂浮垃圾智能识别与自动化清理提供了实验基础。

Abstract: To address the low efficiency, high cost, and operational risk of manual cleanup of offshore floating debris, this paper designs an intelligent detection and cleanup path planning system based on Mask R-CNN. The system uses the Roboflow marine debris dataset, containing 11 categories, 10,000 images, and 56,272 annotated instances. A Mask R-CNN model with a ResNet-50-FPN backbone is adopted for object detection and instance-level region segmentation, while the cleanup route is modeled as an open-loop Traveling Salesman Problem and optimized by a nearest-neighbor greedy algorithm combined with 2-opt local search. Experimental results show that the model achieves a bounding box AP of 53.50% and a mask AP of 52.05% on the validation set. For 200 targets, the path planning algorithm takes about 181.8 ms and reduces the route length by about 10.5%. A dual-mode visualization system with Web and desktop clients is also developed, forming a prototype closed-loop workflow of image input, target detection, path planning, and result display, which provides an experimental basis for intelligent recognition and automated cleanup of offshore floating debris.