基于最佳邻域重构指数的水下高光谱目标检测

Underwater hyperspectral detection method based on optimal neighborhood reconstruction index factor

摘要:

水下机器人仅通过传统光学相机获取图像很难在复杂水下环境中或目标物具有保护色的情况下检测到目标,而通过高光谱技术进行水下目标检测可以改善这一情况。由于直接运用传统高光谱检测方法难以满足水下机器人对水下目标检测的要求,本文提出了一种基于最佳邻域重构指数(ONRIF)的高光谱目标检测方法,该方法通过线性重构的思想进行邻域寻优,选出信息量高且波段相关性低的波段组合,并使用所选波段的融合图像进行目标检测。结果表明,与直接对原始水下海产品高光谱图像进行检测相比,该方法在保证检测效果的前提下,大量减少了检测时间和数据冗余程度。本文还提出了一种在相同环境下对同类目标物的单波段快速采集检测方法,大大提高了采集数据的速度,可以满足水下机器人对海产品检测的需求。

Abstract:

It is difficult for underwater robots to detect targets in complex underwater environments or when the targets have protective colors only by acquiring images with conventional optical cameras, and underwater target detection by hyperspectral techniques can improve this situation. Since it is difficult to meet the requirements of underwater robot for underwater target detection by directly applying traditional hyperspectral detection methods, this paper proposes a hyperspectral target detection method based on Optimal Neighborhood Reconstruction Index Factor (ONRIF), which is based on the idea of linear reconstruction for neighborhood finding, selecting a combination of bands with high information content and low band correlation, and using the band fusion map for target detection. The results show that compared with the direct detection of the original hyperspectral images of underwater seafood, the method substantially reduces the detection time and the degree of data redundancy while ensuring the detection effect. This paper also proposes a single-band fast acquisition detection method for similar targets in the same environment, which greatly improves the speed of data acquisition and can meet the needs of underwater robots for seafood detection.