Högskolan i Skövde

This study investigates how selected hyperparameter configurations affect the performance of a convolutional neural network when classifying LEGO bricks across different image datasets. The study was conducted using two LEGO brick image datasets. Dataset 1 contained rendered and photographic images and was used to evaluate different configurations of image size, batch size, learning rate, and class weights. Dataset 2 consisted of high-resolution photographs captured by the authors and was used to investigate how image size affected CNN model performance when the original image resolution was higher and more consistent.