论文部分内容阅读
高空间分辨率遥感影像通常具有数据量大、背景复杂及地物占比较少等特点。如果直接将RCNN模型应用于高空间分辨率遥感影像目标识别,计算量大且效率低。级联AdaBoost算法识别率高、速度快,但又会产生较多的虚假目标。本文结合RCNN模型和级联AdaBoost算法,提出了一种由粗到精的飞机目标识别方法。首先使用基于HOG特征的级联AdaBoost算法快速提取飞机目标候选区域,然后利用基于卷积神经网络特征的SVM对飞机目标候选区域进行精细识别。试验表明,本文提出的方法在保证准确率的同时,还有效提高了计算效率。
High spatial resolution remote sensing images usually have the characteristics of large amount of data, complex backgrounds and less features. If the RCNN model is directly applied to target recognition of high spatial resolution remote sensing images, it is computationally intensive and inefficient. Cascaded AdaBoost algorithm recognition rate, speed, but will produce more false targets. Based on the RCNN model and the cascaded AdaBoost algorithm, this paper proposes a coarse to fine aircraft target recognition method. Firstly, the candidate target region of aircraft is extracted rapidly by using the cascaded AdaBoost algorithm based on HOG feature, and then the candidate target region of the aircraft is finely identified by SVM based on the features of convolutional neural network. Experiments show that the proposed method can effectively improve the computational efficiency while ensuring the accuracy.