| 基于改进Faster R-CNN的海参目标检测算法 |
| |
| 引用本文： | 熊海涛,林琪,宣魁,葛凤丽,李娟.基于改进Faster R-CNN的海参目标检测算法[J].农业机械学报,2022,53(S2):204-209. |
| |
| 作者姓名： | 熊海涛 林琪 宣魁 葛凤丽 李娟 |
| |
| 作者单位： | 青岛农业大学;福建省海洋生物增养殖与高值化利用重点实验室 |
| |
| 基金项目： | 国家自然科学基金项目(32073029)、山东省自然科学基金重点项目(ZR2020KC027)、山东省研究生教育质量提升计划项目(SDYJG19134)、〖JP2〗国家留学基金项目(201908370048)和福建省海洋生物增养殖与高值化利用重点实验室开放课题(2021fjscq08) |
| |
| 摘 要： | 海参目标检测是实现海参自动化捕捞的前提。为了解决复杂海底环境下背景和目标颜色相近以及遮挡导致的目标漏检问题，本文在Faster R-CNN框架下，提出了Swin-RCNN目标检测算法。该算法的骨干网络采用Swin Transformer，同时在结构上融入了多尺度特征提取层和实例分割功能，提高了算法的自适应特征融合能力，从而提高了模型在复杂环境下对不同尺寸海参的识别能力。实验结果表明：本文方法对海参检测的平均精度均值（mAP）达到94.47%，与Faster R-CNN、SSD、YOLO v5、YOLO v4、YOLO v3相比分别提高4.49、4.56、4.46、11.78、22.07个百分点。
|
| 关 键 词： | 海参 目标检测 目标识别 Faster R-CNN Swin Transformer 深度学习 |
| 收稿时间： | 2022/6/10 0:00:00 |
Sea Cucumber Object Detection Algorithm Based on Improved Faster R-CNN |
| |
| XIONG Haitao,LIN Qi,XUAN Kui,GE Fengli,LI Juan.Sea Cucumber Object Detection Algorithm Based on Improved Faster R-CNN[J].Transactions of the Chinese Society of Agricultural Machinery,2022,53(S2):204-209. |
| |
| Authors: | XIONG Haitao LIN Qi XUAN Kui GE Fengli LI Juan |
| |
| Institution: | Qingdao Agricultural University;Fujian Key Laboratory of Marine Aquaculture and High-value Utilization |
| |
| Abstract: | Sea cucumber object detection is the premise of realizing automatic fishing of sea cucumber. To solve the problem of missed object detection caused by occlusion and the color similarity between object and background in the complex seabed environment, Swin RCNN object detection algorithm was proposed under the framework of Faster R-CNN. The backbone network of the algorithm adopted the Swin Transformer, and the multi-dimensional feature extraction layer was integrated into the structure, which improved the adaptive feature fusion ability of the algorithm and improved the object recognition ability of the model for the different sizes of objects under occlusion in complex environments. The actual experimental results showed that the mean average precision achieved 94.47% for the detection of sea cucumbers by the proposed approach, which was increased by 4.49 percentage points, 4.56 percentage points, 4.46 percentage points, 11.78 percentage points, and 22.07 percentage points compared with Faster R-CNN, SSD, YOLO v5, YOLO v4, and YOLO v3, respectively. The research result had certain reference significance for object detection in other complex environments. Therefore, the study of sea cucumber object detection algorithm in complex seabed environment had important theoretical and application value, and also had guiding significance for intelligent identification of other marine products. |
| |
| Keywords: | sea cucumber object detection object recognition Faster R-CNN Swin Transformer deep learning |
|
| 点击此处可从《农业机械学报》浏览原始摘要信息 |
| 点击此处可从《农业机械学报》下载免费的PDF全文 |
