Classification and object detection on two dimensional health, agriculture, and occupational safety images

Deep learning technologies, especially convolutional neural networks, provide high accuracy in classification and object detection on images. Digitalization has begun in many areas with the progress of artificial intelligence, which includes machine learning and deep learning. In addition, the availability of big data sources has paved the way for the improvement of artificial intelligence and the widespread use of smart systems. This dissertation study focuses on the increase in the use of artificial intelligence in many sectors of human life. In this context, methods are proposed to be used in three different sections. In the first study section, a deep learning-based approach is presented to perform disease detection from tomato leaves. In the second study section, X-ray images are used to classify the relevant image as Covid-19, normal, or pneumonia. A CNN-based hybrid method is proposed to perform the classification. In the last study section, an object-detection approach is discussed in order to determine the helmet-wearing situation to ensure occupational safety, especially in construction areas. This approach is proposed based on the known one-step object detection algorithm Yolov5. Each of the implemented applications within the scope of the dissertation is validated with one or more open-access datasets. The proposed methods are based on convolutional neural networks and successful results have been obtained. This study made a contribution to the literature with applications made with specific data on the improvement of methods. Software tools Matlab, and python was used to implement the proposed methods. Classification and object detection performances were evaluated via used the metrics. Moreover, the proposed methods were compared with similar studies in the literature.