2nd ICAI 2022

International Conference on Automotive Industry 2022

Mladá Boleslav, Czech Republic

the extraction and classification of features in a single step. Other object detection algorithms, such as RCNN, Faster R-CNN, and others, construct probable bounding boxes in an image before running a classifier on them. YOLO v3 uses an architecture of Darknet-53 new neural network for performing feature extraction based on 53 convolutional layers. YOLO v3 makes over scale predictions; it uses three different scales. Prediction over three unique scales for all positions of the input image is the most powerful feature of YOLO v3. System predicts the same way as feature pyramid network, FPN, does. A boundary box, objectness, and class score are all used to make each prediction (Menon, Omman & Asha 2021). The loss function for YOLO v3 in equation (1) is sum of squared error (SSE) loss of

regression, confidence, and classification loss (Redmon et al. , 2016). Taking into account the above, with the use of free tools, the implementation of the architecture presented in Figure 4 was carried out using 2204 picture dataset.

112