| 基于YOLOv5-R-CNN 的海域空间自然资源立体化分布仿真分析 |
| Stereo Distribution Simulation Analysis of Marine Spatial Natural Resources Based on YOLOv5-R-CNN |
| |
| DOI：doi：10.3969/j.issn.1003-2029.2026.02.010 |
| 中文关键词: YOLOv5-R-CNN 海域空间 立体化分布 自然资源 图像增强 |
| 英文关键词:YOLOv5-R-CNN marine space three-dimensional distribution natural resources photographic enhancement |
| 基金项目: |
|
| 摘要点击次数: 254 |
| 全文下载次数: 92 |
| 中文摘要: |
| 由于海域空间环境复杂，获取的海域内水下图像往往会出现雾化、色彩偏差、边缘模
糊等问题，进而影响分析结果的准确性，为此，本文将YOLOv5 （You Only Look Once version 5）和区域卷积神经网络（Region-based Convolutional Neural Network，R-CNN） 融合，建立了基于YOLOv5-R-CNN 融合模型的海域空间自然资源立体化分布仿真分析方法。借助R-CNN 模型结合光波削减规律，去除原始海域空间图形的雾化情况，并校正图形颜色，以提升海域内水下图像质量。引入缓存资源分配机制（Cache Resource Allocation Mechanism，CRAM） 通道融入空间注意机制，自适应学习YOLOv5 多通道权重信息，通过YOLOv5 快速筛选海域内自然资源潜在目标，构建YOLOv5 与R-CNN 的融合模型。利用该模型识别定位自然资源，并依据其空间分布特征建立海域视平面网格，结合闵可夫斯基差理念判断点集交集，计算色调映射平均光照，构建海域空间自然资源仿真立体化分布模型。基于该模型分析自然资源多层次分布，利用R-CNN 预测框识别资源类型，通过平方误差距离立体匹配确定资源边界框，实现对海域空间自然资源立体化分布的仿真分析。实验结果表明：该方法分析的自然资源预测框与实际结果之间差距不足5 cm，准确分析了海域空间自然资源立体化分布情况，为海洋资源开发提供可靠依据。 |
| 英文摘要: |
| Due to the complex spatial environment of the sea area, the obtained underwater images in the sea area often have problems such as atomization, color deviation, edge blur and so on, which will affect the accuracy of the analysis results. Therefore, this paper proposes a simulation analysis method for the three-dimensional distribution of natural resources in the sea area based on You Only Look Once version 5-region based convolutional neural network (YOLOv5-CNN). With the help of R-CNN model and light wave reduction law, the atomization of the original sea area spatial graphics is removed, and the graphics color is corrected, so as to improve the quality of underwater image in the sea area. The cache resource allocation mechanism (CRAM) channel is introduced into the spatial attention mechanism to adaptively learn the multi-channel weight information of YOLOv5. The potential targets of natural resources in the sea area are quickly screened through YOLOv5, and the fusion model of YOLOv5 and R-CNN is constructed. The model is used to identify and locate natural resources, and the view plane grid of the sea area is established according to its spatial distribution characteristics. Combined with Minkowski difference concept, the intersection of point sets is judged, and the average illumination of tone mapping is calculated to build a three-dimensional distribution model of marine spatial natural resources simulation. Based on the model, the multi-level distribution of natural resources is analyzed. The R-CNN prediction frame is used to identify the type of resources, and the resource boundary box is determined by the square error distance stereo matching, so as to realize the simulation analysis of the three-dimensional distribution of natural resources in sea space. The experimental results show that the gap between the prediction frame of natural resources analyzed by this method and the actual results is less than 5 cm, which accurately analyzes the three-dimensional distribution of natural resources in the sea area, and provides a reliable basis for the development of marine resources. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
