Improved YOLOv3-Based Image Recognition of Garbage Plastic Bottles
Weiyi Zheng, Song Cui, Mujun Xie, Ye Wang, Zhong Zheng, Heyu Bian · 2023
Refining waste sorting practices for enhanced garbage recycling rates holds the potential to safeguard the environment while conserving valuable resources. With the advancement of deep learning, the YOLO (You Only Look Once) algorithm has emerged as a promising tool for recognizing beverage bottles amidst garbage. To elevate target detection precision within intricate environmental scenarios, an enhanced approach rooted in YOLOv3 is introduced.This study proposes two pivotal enhancements to augment the capabilities of YOLOv3. Initially, the conventional LeakyReLU activation function is substituted with the Mish function. This replacement leverages Mish's attributes of smoothness and a defined lower bound, surpassing the traits of LeakyReLU. Subsequently, the SE’ASPP module is seamlessly integrated into the convolutional layer. This strategic integration amplifies the network's perceptual scope, thereby fortifying its competence in assimilating multi-scale contexts. This augmentation ultimately propels the model's precision to unprecedented heights.Validation procedures, conducted using a bespoke dataset, substantiate the efficacy of the enhanced YOLOv3 model, which attains an accuracy milestone of 85.26%. This accomplishment signifies a notable enhancement of 2.19% in contrast to the original model's performance.