Children's Drawing Psychological Analysis using Shallow Convolutional Neural Network

Abstract

With the continuous development of civilization, people are paying more attention to mental health issues. Artificial intelligence (AI) technology assessed children's mental health is one of the hot topics in the related field. Drawing test is one of the mainstream psychological assessment methods, and the difficult problem is that how to use computer-assisted psychological analysis to achieve reliable results. Two challenges retain in the literature. First, drawings are different from natural images, which lack information factors such as color and textures, making it difficult for AI algorithms for psychological analysis. Second, the drawings reflect multiple mental health problems. The AI model has to discover hidden factors through a small amount of information displayed on the drawing. This paper proposes a shallow convolutional neural network for feature extraction of children's drawings, combining with the sigmoid function in the fully connected layer for multi-label classification. The classification results shown in the experimental results section justify that the proposed shallow CNN model can assist psychologists in psychological analysis from children's drawings.