Detection of Multi-Class Hazard Signs and Signs Using YOLO Model

The detection of hazard signs and signs used to make the environments where various substances and processed raw materials that may pose a risk during transportation and storage and may pose a risk when found together, safe is of great importance in daily life. It is known that especially employees with different education levels and citizens who are not knowledgeable about the subject have difficulty in recognizing and understanding these signs correctly and cannot have expert-level knowledge about the relevant safety measures due to this deficiency. This situation can lead to unwanted accidents and loss of life and property resulting from incorrect interventions. Therefore, it has become essential for people with all education levels to recognize hazard signs. This study aims to find a solution to this problem by correctly classifying 15669 images consisting of hazard signs and signs in the dataset named "hazmat placard image classification". The dataset we use consists of images consisting of hazard signs and signs and the labels of these images. There are 38 hazard sign and sign classes in the dataset. In our study, image classification was performed using the YOLO_V8 deep learning architecture. The performance of the experimental studies was analyzed using the map50 metric since multi-class estimation was performed. Compared to the studies with few classes in the literature, 38 different classes were classified with the map50 metric value of $88.1 \%$. As a result, the detection of danger signs and boards was performed with high accuracy and fast and accurate detection was achieved before possible accidents and property losses.