为实现货车自动检测记录系统，需要根据货车图象检测进站货车的车锁是否存在，为此，提出了一种基于RBF神经网络和改进遗传算法的货车车锁检测方法，该方法首先提取图象的投影特征，边缘图象的线性短特征以及灰度直方图特征，然后用RBF神经网络进行检测和定位，同时引入遗传算法，利用改进后的遗传算法的高并行性和鲁棒性，可以较快地完成全局搜索，而不会陷入局部最优，实验表明，该方法能有效克服车锁种类多，变形大，以及光照变化的影响，具有较高的速度和检测成功率，能满足于实际应用。

To realize automatically detecting and recording system of freight trains, it's necessary to detect whether the locks of freight trains exist by the image of freight trains when they are entering railway stations. An approach of fast lock detection is put forward, which is based on RBF neural network and improved genetic algorithm. In this method, firstly extract the projection features of images, linear moment features of edge images, and the features of gray histograms of images, and these features reflect different characters of targets from different points of view, and then they are normalized to the input vector of RBF neural network, and RBF neural network is used for detection and location. At the same time, improved genetic algorithm is used to search the whole image, and searching process is speeded up because the genetic algorithm is a parallel and robust algorithm. In addition, image preprocessing is not done alone, and gray variation of images is eliminated during the process of feature extraction. Experiment results show that the method can overcome the problems of many types, deformation, and the variation of environmental brightness, and have the fast speed and high success rate of detection, and can be put into practical application. Therefore, the method has significant engineering value.