LEYOLO OBJECT DETECTION USING CNN WITH ADVANCED FEATURES

Object detection, being a major problem in computer vision, has a variety of applications that include selfdriving cars, surveillance, and medical diagnostics. In this paper, we discuss LeYOLO, an updated version ofthe existing LeYOLO model, with several significant changes which increase its efficiency, scalability, androbustness. Multi-scale feature extraction, lightweight architecture, small and far object detection, hybridactivation functions, and an adaptive loss function are our proposed changes. These improvements eliminate theshortcomings present in the original LeYOLO model, resulting in more accurate detections and a fasterinference speed. Testing was carried out on the COCO dataset, where we showed tremendous gains inperformance in favor of our model against the baseline model.