基于深度学习与图像处理的废弃物分类与定位方法

针对现有人工垃圾分类工作环境恶劣、自动化程度差等问题,提出基于深度学习与图像处理的废弃物分类与定位方法,为智能分拣提供理论依据.建立基于Inception模块与残差单元,搭建改进的卷积神经网络废弃物分类模型,预测目标物体种类.针对复杂环境采集到的图像,利用图像处理算法对图像降噪、阈值分割、边缘检测,有效提取目标轮廓信息,并结合质心定位算法实现废弃物准确定位.实验结果表明:该方法中废弃物分类模型预测准确率可达88.8％,基于轮廓信息的质心定位算法可以准确定位目标,具备较强的废弃物分类与定位能力.

Research on Waste Classification and Location Method Based on Deep Learning and Image Processing

In view of the poor working condition and ineffective automation environment in current waste treatment, a waste sorting and detecting method based on deep learning and image processing is proposed here in an attempt to provide a theoretical basis for intelligent sorting. The proposed theoretical framework is grounded on the Inception module and residual unit to establish an improved convolutional neural network model for sorting the waste and detecting the types of target objects. The images collected in complex environments are treated using image processing algorithms in order to reduce image noise, segment threshold, and detect the edges of the target objects. This method is proved to enable an effective extraction of the contour information from the target objects, and with a combination with the centroid locating algorithms, an effective waste positioning is achieved. An accuracy rate of 88% in terms of waste classification is obtained in the experiment. This result shows that the centroid locating algorithms based on contour information has sufficient capabilities for waste detection and classification.