Abstract:
Measurement of cognitive development in a child is a very old phenomenon and for many years psychologists used different tools and techniques to measure the intelligence level in a child. In this research, we have used one such technique called Florence Good Enough scoring schema, which is helping medical researchers for many decades to measure the intelligence quotient in children. We have presented the computerized system which automates the process of generating the Florence Good Enough scoring that will assist the psychologists in prescribing the necessary treatment to a child. We have focused on the first three categories Gross detail, Attachment, and Head detail which contains a total of 9 classes. We have collected hand-drawn images from children of age varying from 7 to 12 years old and labeling those images with the help of a clinical expert. Applied CNN-based multi-label classification method with pre-trained models and two-step fine-tuning technique to determine the existence of components in an image i.e. head, trunk, legs, arms, arms & legs attached to the trunk, arms & legs attached to trunk at the correct point, eyes, mouth, and hair. Define three models for Gross detail, Attachment, and Head detail and evaluate the results for each individual category as well as compares the performance of all the experiments. Amongst all the techniques sketch-based twostep fine-tuning method produced remarkable results where Gross detail achieves 81.1%, Attachment 99.2%, and Head detail achieves 79.3% accuracy. Finally, applied ANOVA test to statistically compares the performed of the experiments with each other to see if there is any significant difference between the networks or not
