A Lightweight YOLOv4-SVM Model for Automated Waste Monitoring in Smart Cities

Waste at the municipal level has become one of the salient problems of modern urban management practices. Increasing use of products that can be easily thrown away, such as bottles, plastic bags, cans, and straw, puts extensive pressure on sanitary infrastructure. Manual inspection and collection procedures represent not only inefficient, but also unsuitable mechanisms to cope with complexities of denser and larger urban environments. To address this problem, this paper proposes a detection system combining YOLOv4 and a support vector machine (SVM) to improve the accuracy of recognition in a cluttered environment. Four common types of trash are targeted by the system: plastic bags, bottles, tin cans, and straws. Initial detection is performed using YOLOv4, and the SVM additionally filters the detection results to remove false positives. Experiments on a subset of the Open Images dataset demonstrate that the proposed method has satisfactory performance in mean average precision (mAP), precision, and recall. This system has high potential for widespread application in smart city infrastructure and can act as a baseline for future automated waste management systems.