论文部分内容阅读
以三种飞机模型作为待识别目标 ,模拟真实场景 ,对用于多目标分类识别的级联神经网络重新进行了研究。实验发现识别率下降的主要原因是实际采集的目标发生的复杂畸变与计算机模拟产生的效果并不一样。用采集得到的目标图像作为训练样本 ,对网络重新构造和训练 ,取得了好的实验结果。分析了其中涉及到目标定位、图像分割等图像预处理问题。提出了一种基于二值图像形态学腐蚀运算的快速目标检测定位法 ,可快速有效地对目标进行检测定位。
Taking the three aircraft models as the target to be identified and the real scene simulated, the cascade neural network used for multi-target classification and recognition was re-studied. The experiment found that the main reason for the decrease of recognition rate is that the complex distortion of the actual acquisition target is not the same as the result of computer simulation. With the acquired target image as a training sample, the network is reconstructed and trained, and good experimental results have been obtained. The problems of image preprocessing, such as target location and image segmentation, are analyzed. A fast target detection and location method based on binary image morphological corrosion operation is proposed, which can detect and locate the target quickly and effectively.