将车辆检测的过程视为一种“抗体”检测“危险抗原”的过程, 其中车辆是“危险抗原”, 车辆检测模板是“抗体”。利用一些车辆图像作为训练样本, 建立一种抗体网络学习并获取一组优化的“抗体”。这些“抗体”经过与待测影像的匹配, 实现对道路车辆目标的有效提取。采用0.6m分辨率的QuickBird全色数据进行实验, 实验结果验证了该方法的有效性和可行性。
This paper presents an antibody network approach for vehicle detection from high resolution satellite imagery. This approach regards the vehicle detection procedure as a procedure that antibodies recognize danger antigens, where vehicles are “dangerous antigens” and vehicle detection templates are “antibodies”. In this paper, some vehicle images are collected as learning examples, and an antiboby network is proposed to learn optimal “antibodies”, which can be used to detect vehicles through the proposed matching algorithm. Experiments on Quickbird satellite images are given to show the feasibility and performance of the proposed approach.