基于GAICNet的垃圾识别分类检测网络

Garbage identification, classification and detection method based on GAICNet

随着社会经济的飞速发展和人们物质生活水平的不断提高,所产生垃圾的种类和数量也在大幅增加,为了有效提高垃圾资源的回收利用率和垃圾分类智能化水平,本文提出了基于GAICNet的智能垃圾识别分类检测网络。使用华为云人工智能大赛公开的数据集,对训练集使用LabelImage软件进行标注,然后为了扩大训练数据,强化网络模型对普通数据集的泛化能力,设计了一种具有普适性的非线性曲线映射图像增强方法。最终,构建了一个全局感知特征聚合模块与全新的核心自注意力机制的垃圾识别分类网络(Garbage identification and classificationNet,GAICNet)。该轻量级网络是以ResNet18为骨干网络进行多层次的特征提取,实验显示该算法在现有复杂实际场景垃圾分类中的检测准确率(Accuracy)可达到97.3%,具有较高的实用价值和市场前景。

With the rapid development of social economy and the continuous improvement of human living standards,the types and quantity of waste are also greatly increasing.In order to effectively improve the recycling rate of waste resources and the intelligent level of waste classification,an intelligent waste identification,classification and detection network based on GAICNet is proposed in this paper.Using the data set published by Huawei cloud artificial intelligence competition,the training set is labeled with LabelImage software.Then,in order to expand the training data and strengthen the generalization ability of network model to ordinary data sets,a universal nonlinear curve mapping image enhancement method is designed.Therefore,a garbage identification and classification network(GAICNet)with global perceptual feature aggregation module and a new core self attention mechanism is constructed.The lightweight network takes ResNet18 as the backbone network for multi-level feature extraction.The experiment shows that the detection accuracy of the algorithm in the existing complex actual scene garbage classification can reach 97.3%,which has high practical value and market prospect.