本文提出了一种基于卷积神经网络的目标检测算法,解决了传统人工选取特征准确性不足的缺点;同时引用回归思想取代了滑动窗口或者建议区域的遍历式的工作方式,加快了目标检测的速度;同时在训练中加入难例挖掘,提高了训练之后网络的准确性。并进行三种数据集的训练和测试,对数据进行详尽分析。 In this paper, a target detection algorithm based on convolutional neural network is proposed, which solves the shortcomings of the lack of accuracy of the traditional manual selection features. At the same time, the reference of regression theory replaces the traversal work mode of sliding window or proposed region, accelerating the target detection Speed; at the same time join the hard work mining mining, improve the accuracy of the network after training. And three data sets of training and testing, a detailed analysis of the data.