Artificial intelligence-based automatic helmet detection system for occupational safety inspection in agricultural machinery factories

The focus of this thesis is to develop an artificial intelligence-based system for detecting whether individuals working in tractor and agricultural machinery factories are wearing helmets. For this purpose, a model based on transfer learning was developed to create an effective occupational safety application, accurately determining the helmet-wearing status of individuals. This model performs deep feature extraction using nine different pre-trained artificial neural networks. These networks are: (i) MobileNetV2, (ii) ResNet50, (iii) DarkNet53, (iv) AlexNet, (v) ShuffleNet, (vi) DenseNet201, (vii) InceptionV3, (viii) InceptionResNetV2, and (ix) GoogleNet. The feature vectors obtained from these networks were subjected to cyclic neighborhood component analysis (CNCA) for feature selection, and these features were then classified using the k-nearest neighbor (kNN) method. The nine different classification outputs were combined using the cyclic weighted voting (CWV) algorithm to achieve the optimal result. To evaluate the system's performance, an image dataset was created by collecting images from tractor and agricultural machinery factories via the internet. This dataset is divided into two categories: (1) 392 individuals with helmets and (2) 314 individuals without helmets. Consequently, the artificial intelligence-based helmet detection method proposed in this thesis demonstrated high performance with an accuracy rate of 90.39%. The experiments conducted using CNCA and kNN classifiers showed the best results with the DenseNet201 network. The obtained results highlight the potential of artificial intelligence in the field of occupational health and safety. Especially when used in enclosed spaces, such systems can make the tasks of occupational safety experts more efficient and minimize human errors. Key Words: Occupational health, ppe, farm machinery, artificial intelligence, tractor workshops, deep learning, machine learning