Child psychological drawing pattern detection on OBGET dataset, a case study on accuracy based on MYOLO v5 and MResNet 50

The interpretation of drawing psychological data is amazing, and they behave differently in various situations like age. This research compares different deep learning methods and behaviors on children's data. Therefore, after presenting the collected dataset and statistical analysis, two methods, the proposed modified YOLO V5 as MYOLO V5 and ResNet 50 as MresNet 50, are compared and discussed on the OBGET dataset for children. The classification accuracy of the two proposed methods, the proposed MYOLO V5 and MResNet 50 on the OBGET database for children, has been measured. To prepare conditions for comparing these two algorithms for the first time, necessary preprocessing and the proposed semi-automatic labeling were performed on children's OBGET samples. Children's OBGET contains 386 Original Bender Gestalt Drawing Test samples for children. The proposed MResNet 50 and MYOLO V5 classify patterns in children's OBGET samples as single-stage deep learning methods to obtain acceptable pattern detection accuracy. After deploying the proposed methods on the proposed collected dataset, results show that the proposed MYOLO V5 presents higher accuracy.