This study outlines the creation of a visual recognition system designed for the automatic identification and assessment of Personal Protective Equipment in workplace settings. The aim is to reduce workplace risks by ensuring the proper usage of PPE items like helmets, vests, gloves, boots, and eye protection. The process included gathering data, preprocessing steps (which involved creating bounding boxes, extracting regions of interest, resizing, and manual annotation), and training models using six different frameworks: Vision Transformer (ViT), Yolo v11, DeiT, Swin Transformer, VGG19, and CvT. The models were assessed with an 80/20 split between training data and test data, utilizing metrics such as accuracy, precision, recall, and F1 score. After performing optimization through GridSearchCV, ViT emerged as the top model, presenting an accuracy of 99. 6%, precision at 99. 2%, recall at 98. 7%, and an F1 score of 99. 66%. These findings support the efficacy of deep learning frameworks for ensuring compliance with PPE regulations.