| 基于Faster R-CNN的海面垃圾检测研究 |
| |
| 引用本文： | 韦波,张衡,王斐,王书献,杨昱皞,姚宇青,戴阳.基于Faster R-CNN的海面垃圾检测研究[J].环境工程,2022,40(7):153-158. |
| |
| 作者姓名： | 韦波 张衡 王斐 王书献 杨昱皞 姚宇青 戴阳 |
| |
| 作者单位： | 1. 上海海洋大学 信息学院, 上海 201306; |
| |
| 基金项目： | 上海市2020年度"科技创新行动计划"(20dz1206400)上海市青年科技英才扬帆计划(19YF1460000) |
| |
| 摘 要： | 针对当前全球海洋垃圾污染造成严重生态破坏等问题,提出了一种基于改进Faster R-CNN的海面垃圾检测算法以及视频中前后2帧目标是否为同一目标的方法。改进Faster R-CNN算法通过将常用的VGG16特征提取网络替换为ResNet101网络并融入特征金字塔,提高对小目标的检测精度;判断视频前后两帧目标是否为同一目标的方法则通过对比前后2帧目标的面积、重合度以及颜色差异度确定是否为同一目标。在现场拍摄数据上的实验结果表明,与传统Faster R-CNN相比,该改进Faster R-CNN的mAP值提高了4.9%,损失曲线的收敛速度更快,且在实际检测中的检测效果更好;前后两帧是否同一物体的判断方法在九段视频的最高精测判断精度高达100%,平均准确率为93%。该研究方法主要包括以下优点:1)改进Faster R-CNN在海面小目标垃圾检测上具有更高的精度;2)判断视频中前后2帧目标是否为同一目标的算法代码复杂度小,方便根据实际情况更改判断阈值。
|
| 关 键 词： | 改进Faster R-CNN 小目标 特征金字塔 同一目标判别法 |
| 收稿时间： | 2021-09-18 |
RESEARCH ON SEA GARBAGE DETECTION BASED ON FASTER R-CNN |
| |
| Affiliation: | 1. School of Information, Shanghai Ocean University, Shanghai 201306, China;2. Key Laboratory of Innovation in Pelagic and Polar Fisheries, Ministry of Agriculture and Rural Affairs, East China Sea Fisheries Research Institute, Chinese Academy of Fishery Sciences, Shanghai 200090, China |
| |
| Abstract: | Given the serious ecological damage caused by global marine garbage pollution,a sea garbage detection algorithm based on improved fast R-CNN and a method of whether the targets in the front and back frames of the video are the same target were proposed.The improved fast R-CNN algorithm improved the detection accuracy of small targets by replacing the commonly used VGG16 feature extraction network with ResNet101network and integrating it into the feature pyramid;the method to judge whether the two frames before and after the video were the same target,was to determine whether they were the same target by comparing the area,coincidence and color difference of them.The experimental results on the field shooting data showed that compared with the traditional fast R-CNN,the map value of the improved fast R-CNN in this paper was increased by 4.9%,the convergence speed of the loss curve was faster,and the detection effect in the actual detection was better;the proposed method for judging whether the two frames are the same object had a peak precision of 100% and an average accuracy of 93% in nine videos.The proposed method mainly included the following advantages:1) the improved fast R-CNN had higher accuracy in the detection of small target garbage on the sea surface;2) the algorithm code complexity of judging whether the target of the first and second frames in the video is the same target is small,making it convenient to change the judgment threshold according to the actual situation. |
| |
| Keywords: | |
|
| 点击此处可从《环境工程》浏览原始摘要信息 |
|
点击此处可从《环境工程》下载全文 |
|
