基于CNN池化和进化策略的一般神经网络图像分类研究

网络的爆炸式发展产生了海量的图像,图像标签的错误和缺失比较常见,图像分类研究很有必要。CNN池化能够提取到输入矩阵的重要特征,降低数据的维度。进化策略是模仿生物"优胜劣汰"进化方式的一种启发式算法,能快速找到问题的解。本文基于CNN池化提取一组有正确标签的图像的特征,搭建层数为3的神经网络,进化策略优化初始权重,通过训练集训练分类模型,通过测试集来验证模型的优劣,并使最终的模型实现对未知类别图像的高效分类。实例验证阶段收集10类100张犬类图片,按照各研发步骤进行实验,算法结果验证了进化策略优化权重的必要及神经网络模型的高效。

Research on general neural network image classification based on CNN pooling and evolutionary strategy

The explosive development of the network has produced a large number of images,and the errors and missing of image labels are more common,and the research on image classification is necessary.CNN pooling can extract important features of the input matrix and reduce the dimensionality of the data.Evolutionary strategy is a heuristic algorithm that imitates the evolutionary method of"survival of the fittest"of biology,which can quickly find the solution of the problem.Based on CNN pooling,the paper extracts the features of a set of images with correct labels,builds a neural network with three layers,optimizes the initial weights by evolutionary strategy,trains the classification model through the training set,and verifies the pros and cons of the model through the test set.The final model realizes the efficient classification of unknown category images.In the instance verification stage,100 dog pictures of 10 categories are collected,and the experiment is carried out according to the above steps.The algorithm results verify the necessity of the evolution strategy to optimize the weight and the efficiency of the neural network model.