基于目标与环境FD模型的多特征检测算法适应性评估

智能变形、变色、变温、变谱技术发展趋势下，低特征目标加速实现与自然地物背景的特征融合，导致复杂自然背景环境下低散射、微反射、弱辐射目标的检测与评估愈发困难，特定场景下潜在威胁目标的检测方法快速决策与准确评估成为了难题。为了提升离散目标、伪装目标、弱小目标、异常目标等低特征目标与复杂自然背景环境融合场景下的多特征检测算法的选择效率及其检测准确度，提出了目标与背景环境融合度（FD）参数模型，并设计了植被伪装目标嵌入草地背景、植被伪装目标嵌入土壤背景、植被及水泥路伪装目标嵌入土壤背景以及植被、水泥路、土壤伪装目标分别嵌入草地、水泥路、土壤背景等4种不同波谱特征分布场景的模拟图像数据，以及信噪比为200，400与800的高斯白噪声分别加入场景一的3种不同级别噪声比例的模拟图像数据。通过综合目标波谱信息、背景波谱信息、数据噪声比例等多种因素的综合试验分析，开展了基于目标与环境FD模型的多特征检测算法适应性评估研究。结果表明，在标准差均小于0.08的条件下，MtACE，MtAMF，MtCEM，SumACE，SumAMF，SumCEM，WtaACE，WtaAMF，WtaCEM等9大经典多特征检测算法对于4种波谱分布场景检测结果的FD参数平均值分别为0.320 0，0.350 2，0.862 4，0.365 8，0.365 8，0.846 1，0.680 0，0.680 0和0.948 2；在标准差均小于0.07的条件下，9大经典多特征检测算法对于3种不同级别噪声比例数据检测结果的FD参数平均值分别为0.313 5，0.320 9，0.774 7，0.369 6，0.369 6，0.847 5，0.695 6，0.695 6和0.960 3。通过不同波谱分布场景及不同噪声级别条件下的检测与融合度评估试验分析，实现了多特征检测算法的适应性能排序，大幅提升复杂场景下多种低特征目标的检测效率。综合波谱与噪声因素，对于复杂场景下离散分布的低特征目标检测，9大经典多特征检测算法的优先级顺序为：MtACE>MtAMF>SumACE=SumAMF>>WtaACE=WtaAMF>MtCEM>SumCEM>WtaCEM。

Adaptability Analysis of Multiple Features Detection Algorithms Based on Fusion Degree Model Between Target and Environment

Under the development trend of intelligent deformation, color change, temperature change, and spectrum change technology, low-feature targets accelerate the realization of feature fusion with the background of natural features, which makes the detection and evaluation of low-scattering, micro-reflection, and weak-radiation targets under complex natural backgrounds more and more difficult . Furthermore, the detection method, rapid decision-making and accurate evaluation of potential threat targets in certain situations have become difficult problems. This paper has proposed a parameter model for Fusion Degree(FD) between targets and background environments to improve the selection efficiency of multi-feature detection algorithms and the accuracy of the detection effect evaluation under the fusion scene of complex natural background environment with low-feature targets, such as discrete targets, camouflage targets, small targets, abnormal targets and so on. At the same time, simulated image data of 4 different spectral feature distribution scenes were designed, including vegetation camouflage targets embedded in the grass background, vegetation camouflage targets embedded in the soil background, vegetation and cement road camouflage targets embedded in the soil background, and vegetation, cement road, and soil camouflage targets embedded in the grass, cement road, and soil background respectively. Furthermore, signal noise ratio(SNR) of 200, 400 and 800 were applied to the spectral feature distribution scenes in which vegetation camouflage targets were embedded in the grass background. Through comprehensive Testal analysis of multiple factors such as spectrum information of targets, spectrum information of background, data noise ratio, etc., the research on threat evaluation of multi-feature detection algorithm was carried out, which was based on FD model between target and environment. Under the condition that the standard deviation is less than 0.08, the average values of FD parameters of the 9 classic multi-feature detection algorithms such as MtACE, MtAMF, MtCEM, SumACE, SumAMF, SumCEM, WtaACE, WtaAMF, and WtaCEM for the detection results of the 4 spectrum distribution scenes are 0.320 0, 0.350 2, 0.862 4, 0.365 8, 0.365 8, 0.846 1, 0.680 0, 0.680 0, 0.948 2, respectively. Meanwhile on the condition that the standard deviation is less than 0.07, the average values of FD parameters of the 9 classic multi-feature detection algorithms for detection results of 3 different levels of noise ratio data are 0.313 5, 0.320 9, 0.774 7, 0.369 6, 0.369 6, 0.847 5, 0.695 6, 0.695 6, 0.960 3, respectively. In this paper, through the analysis of detection and fusion evaluation tests under different spectrum distribution scenarios and different noise levels, the threat level ranking of multi-feature detection algorithms is realized, and the detection efficiency of multiple low-feature targets in complex scenarios is greatly improved. Considering spectrum and noise factors, for the detection of discretely distributed low-feature targets in complex scenes, the priority order of the 9 classic multi-feature detection algorithms is: MtACE>MtAMF>SumACE=SumAMF>>WtaACE=WtaAMF>MtCEM>SumCEM>WtaCEM.