为提高光电系统对弱小目标的识别和分类能力,降低算法对硬件平台和数据的依赖,提出一种无监督分类方法-基于目标深度特征聚类的细粒度分类方法。该方法通过轮廓、颜色、对比度等浅层特征提取提示目标,经超分辨处理后,利用卷积神经网络对目标的深层特征进行编码,进一步采用基于注意机制的主成分分析方法进行降维生成表征矩阵,最后利用聚类的方式实现目标细粒度分类。实验验证了基于不同神经网络的深度聚类方法在不同数据集上的分类性能,其中采用ResNet-34聚类方法在CIFAR-10测试集上细粒度分类性能达92.71%,结果表明,基于深度聚类的目标细粒度方法能够取得与强监督学习方法相当的目标分类效果。此外,还可以根据不同簇数和聚类等级的选择实现不同细粒度的分类效果。
目标定位是光电转塔典型功能和任务之一,对其定位精度的考量也是转塔作战技术指标之一,针对该问题,从理论和仿真角度进行了分析.分析目标定位中用到的坐标系及其相互转换关系,给出光电转塔视轴反演、有源目标定位、无源目标定位的算法流程,通过仿真实验加以验证,考虑了定位过程中可能的随机误差来源,并分析是否采用均值滤波及其对定位结果的影响,最后通过Monte-Carlo分析计算了定位精度.分析结果表明:1)有源定位比无源定位的精度高,在仿真假设条件下,精度约提高1倍;2)均值滤波后,定位精度有较大提升(约提高15倍);3)18 km距离时典型无源定位精度在80%置信度条件下约为39.4 m;4)统计直方图反映出80%置信度CEP半径及最大误差距离随载机位置、姿态、转塔视轴等(体现在目标载机距离上)不同参数的变化结果.
Based on the analysis of the existing image and spatial data fusion methods, a new method of integrating real-time image and 3d digital map of photoelectric sensor in the airborne field is proposed. This method can realize the fusion of 2d photoelectric images into 3d terrain space, and the combination of real-time high resolution photoelectric images and digital maps can directly reconstruct the surrounding environment of the aircraft and reduce the workload of pilots, or facilitate the analysis process of intelligence analysts. It has certain application potential in precision strike field.
飞行高度精确测控是双模导引头执行目标指示、航姿遥测的主要实施单元.针对传统高度测量方法的缺陷和导引头趋于小型化、智能化、标准化的要求,设计并实现了一种具有大气压采集、温度补偿、大气压高度转换与经纬度信息实时同步的高度定位系统,其大气压高度转换率高、输出数据可靠稳定.经过外场飞控综合性能测试和模拟打靶功效测试验证,表明高度定位系统测量数据精准度和实时性满足导引头研发需求.
Introduces the image and spatial data fusion in photoelectric detection application. Focus on the optimization of 3D scene reconstruction. In airborne imaging applications, aiming at the problems of massive terrain data, this paper proposed a dynamic data scheduling strategy which is based on state-tree from simplification, and present a terrain data dynamic schedule framework from render optimization. For the suggested optimized procedure and framwork, give a experiment and couclusion based on programmimg, it prove that the suggested dynamic schedule strategy in this paper could fastly construct three-dimensional scene in flight simulation, could speed up the three-dimensional visulization, it could meet the practical requiremnet of engineering in flight simulation.