Author(s)

Md. Rakib Jabed, Md. Shamsuzzaman

Department of Computer Science and Engineering, Begum Rokey University, Rangpur, Bangladesh.

In order to safeguard the biological framework on which human creatures depend for presence and make society feasible improvement, it is becoming increasingly important to classify garbage. In any case, individuals are not recognizable with the classification strategy, so it is troublesome for individuals to accurately get the classification of each kind of rubbish. A proper waste management system is a primary task in building a smart and healthy city. In arrange to direct individuals to classify garbage accurately, this paper proposes a strategy of rubbish classification and acknowledgment based on YOLOv7 which is a cutting edge real time object detector. Performance of this model is compared along with two other object detectors where Mask-RCNN achieved f-measurement of 85%, YOLOv5 achieved f-measurement of 95.1% and YOLOv7 achieved f-measurement 95.9%. We have used non-decomposable multiclass garbage images which entails messy backgrounds with unwanted images as well. Four classes of non-decomposable garbage data namely chips packet, plastic bottle, polythene and image with multiclass garbage with 1000 images are prepared for our dataset. Our experimental models performed well in classifying garbage images with cluttered backgrounds. We compared our test results to previous studies in which the majority of the models were tested and trained using laboratory images. The test comes about illustrates that the classification framework features a sensible degree of accuracy and the segmentation recognition impact is way better within the case of point-by-point picture, which can proficiently and helpfully total the rubbish classification errand.

Mask-RCNN, YOLOv5, YOLOv7, Garbage Classification, Machine Learning

Jabed, M. and Shamsuzzaman, M. (2022) YOLObin: Non-Decomposable Garbage Identification and Classification Based on YOLOv7. Journal of Computer and Communications, 10, 104-121. doi: 10.4236/jcc.2022.1010008.