Small Visual Object Detection in Smart Waste Classification Using Transformers with Deep Learning

Abstract

AbstractSmart object waste classification is relatively essential for protecting the environment and saving resources. This is considered a vital pathway towards sustainability. In waste classification, we see that it is challenging to detect waste of small visual objects with low resolutions that directly affect the overall performance of waste classification. While current visual object detection algorithms focus on the exploration of larger objects, the development of small object detection is being expanded relatively slowly due to the inability to acquire more visual information. In this paper, we propose a novel method combining contextual information and multiscale learning to improve small object detection performance in waste classification by enabling small object detection to obtain more feature information at high resolution. Furthermore, based on the advantages of parallel computing in Transformers, we utilize the DETR model to explore our method. The experimental results show that our method achieves high accuracy in the detection of a small object in waste.KeywordsSmall object detectionTransformersWaste detectionWaste classification