针对传统卷积神经网络(CNN)模型构建过度依赖经验知识、参数多、训练难度大等缺点,同时鉴于复杂多类问题的CNN模型构建策略的重要价值,提出一种自适应深度CNN模型构建方法.首先,将初始网络模型的卷积层和池化层设置为仅含一幅特征图;然后,以网络收敛速度为评价指标,对网络进行全局扩展,全局扩展后,根据交叉验证样本识别率控制网络展开局部扩展,直到识别率达到预设期望值后停止局部网络学习;最后,针对新增训练样本,通过拓展新支路实现网络结构的自适应增量学习.通过图像识别实验验证了所提算法在网络训练时间和识别效果上的优越性. In view of the disadvantages of overconforming traditional convolution neural network (CNN) model, such as relying too much on empirical knowledge, parameters and training difficulties, and considering the important value of constructing CNN model for complex multi-class problems, this paper proposes an adaptive depth CNN model building method First, the convolutional layer and the pooling layer of the initial network model are set to only contain one feature map. Then, the network is globally expanded according to the convergence rate of the network, and then, based on the cross-validation sample identification rate The control network expands locally until the recognition rate reaches the preset expected value, and stops the local network learning. Finally, the new incremental branch is used to realize the adaptive incremental learning of the network structure by adding new training samples. The Superiority of Algorithm in Network Training Time and Recognition.