Automated Grading and Classification of Hand-Drawn Sketches Using Deep Learning

Automated evaluation systems have the potential to greatly simplify the assessment process in educational environments. However, traditional automated evaluation systems often fall short when assessing hand-drawn sketches, a prevalent component of student responses. They struggle to effectively assess hand-drawn sketches, particularly those exhibiting diverse color variations and shapes. This research introduces a deep learning-based approach to address this limitation. The system effectively executes multi-class classification and quality assessment of handdrawn images using the VGG16 convolutional neural network (CNN) architecture. Trained on a dataset labeled by human experts, the model learns to extract patterns and features from sketches, enabling accurate categorization into predefined classes and reliable grade prediction. Experimental results demonstrate exceptional performance, achieving an accuracy of 96.60% for label classification and 88% for grade assessment. These findings underscore the potential of deep learning for automating the assessment process in educational settings, providing a more efficient and objective evaluation of student work.