基于ResNet-50垃圾分类算法的改进及应用

Research on garbage classification algorithm based on improved ResNet-50

随着人们生活水平和消费水平的不断提高,垃圾问题日益严峻。针对当前垃圾分类易出错、准确率低等问题,本文提出了一种改进的ResNet-50识别算法,首先通过二维Gamma函数对图像进行光照校正预处理;然后,采用Leaky ReLU激活函数,并把激活函数和BatchNormalize层的位置放在了卷积神经网络的卷积操作之前,优化了ResNet-50网络结构。最后,收集常见的4种类型垃圾进行训练、测试得到最优网络模型。经实验验证,该模型的准确率达到99%,识别效果较佳。为营造共建共享氛围,实现垃圾快速有效分类,推动绿色生活方式提供了理论依据。

With the continuous improvement of people’s living standards and consumption levels, the garbage problem is becoming increasingly serious. Aiming at error-prone and low-accuracy problems in garbage classification, this paper proposes an improved ResNet-50 recognition algorithm. Firstly, the image is pre-processed by two-dimensional Gamma function;then, the Leaky ReLU activation function is used, and the positions of the activation function and BatchNormalize layer are placed before the convolution operation of the convolution neural network to optimize the ResNet-50 network structure. Finally, four types of common garbage are collected for training and testing to obtain the optimal network model. The experimental results show that the accuracy of the model reaches 99% and the recognition effect is better. It provides a theoretical basis for creating a co-construction and sharing atmosphere, realizing rapid and effective waste classification, and promoting a green lifestyle.