基于背景分割法的数码迷彩设计及伪装性能评估

Digital camouflage design and camouflage performance evaluation based on background segmentation method

针对自然环境中作战人员及军事设备的伪装隐蔽性问题,提出了一种基于背景分割法的数码迷彩设计方法。首先,通过分析环境场景的特征,采用特征提取技术,确定最佳的迷彩图案元素。其次,利用背景分割法进一步细化数码迷彩图案的色彩分布占比,使数码迷彩图案置于环境场景的任意位置都能保持良好的伪装隐蔽效果。最后,运用ResNet-50深度神经网络模型和Mask R-CNN迷彩目标分割模型对基于背景分割法设计得到的数码迷彩图案进行伪装隐蔽性能评价。结果表明,相比于传统迷彩设计方法,此方法得到的数码迷彩图案的伪装隐蔽性能显著提升,隐蔽融合度均值提高至99.00%以上,环境场景的相似度提高20.41%,平均分割识出率降低82.21%。

A digital camouflage design method based on background segmentation was proposed to address the camouflage concealment issues of personnel and military equipments in natural environment.Firstly,by analyzing the characteristics of environmental scenes and employing feature extraction techniques,the optimal elements of camouflage pattern were determined.Secondly,the color distribution ratio of digital camouflage pattern was further refined using background segmentation,ensuring that the digital camouflage pattern could maintain effective camouflage concealment regardless of its position in the environmental scene.Finally,the deep neural networks model ResNet-50 and camouflage target segmentation model Mask R-CNN were utilized to evaluate the camouflage concealment performance of digital camouflage patterns designed based on background segmentation.The results indicated a significant improvement in the concealment effectiveness of digital camouflage compared to traditional camouflage generation method.The mean fusion fitness exceeded 99.00%,the environmental scene similarity increased by 20.41%,and the mean pixel-wise true positive decreased by 82.21%.