Abstract

Disciplinary violations occurred in the SDP Palembang Transportation Polytechnic environment, the violations were in the form of many cadets who smoked. This study aims to provide alternative solutions by utilizing video image data obtained from CCTV to develop CNN and YOLO models to detect smoking violations by cadets in the SDP Palembang transportation polytechnic environment. The benefits of research in the scientific field by testing the level of accuracy of the model developed using the YOLO and CNN algorithms on CCTV image data. The results showed that the precision value increased along with the number of epochs that took place, where the final precision value obtained was at a value of 0.90152, while the recall value showed 0.82878. The CNN algorithm Recall value showed 0.9625, while the CNN algorithm Precision value showed 0.7349. The HOG algorithm Precision value showed 0.86562, while the HOG algorithm Recall value showed 0.6715. The combined recall value of the CNN and YOLO algorithms showed 0.94339 while the combined Precision value of the CNN and YOLO algorithms showed 0.89285. The model has a Mean Average Precision (mAP) value of 0.89109, which shows that the model that has been developed has a good level of accuracy.Keywords: Accuracy; CNN algorithm; Mean Average Precision (mAP); Precision value; YOLO algorithm