To solve the problems of over-fitting, poor convergence, and reduced recall and accuracy of traditional image recognition algorithms, a junk image recognition algorithm based on deep learning is proposed. Dropout is introduced to overcome over fitting, adagrad adaptive method is used to debug the parameters of deep neural network, and ReLU is adopted to solve the gradient dispersion of neural network training, realize the centralized processing of garbage image data, and extract the edge features, color features and texture features of garbage image in the data set, respectively, shape features. The modified probability density function is used for image classification. According to the different characteristics of garbage image, the image to be recognized is divided into different categories to complete garbage image recognition. The experimental results show that the designed algorithm has good convergence, high recall and accuracy, and short recognition time, indicating that the algorithm is feasible and practical.

针对传统图像识别算法过拟合、收敛性差、召回率和准确率降低等问题，提出了一种基于深度学习的垃圾图像识别算法。引入Dropout克服过拟合，采用adagrad自适应方法调试深度神经网络参数，采用ReLU解决神经网络训练梯度分散，实现垃圾图像数据集中处理，提取边缘特征，数据集中垃圾图像的颜色特征和纹理特征，分别是形状特征。修改后的概率密度函数用于图像分类。根据垃圾图像的不同特征，将待识别图像划分为不同类别，完成垃圾图像识别。