In underground mining operations, blast-induced rock fragmentation provides information that can be useful for downstream ore handling operations. Continuous information on rock fragmentation is essential to ensure efficient loading operations using semi-automated and automated LHDs. The manual work of classifying fragmentation is a laborious and time-consuming task. This paper presents an early stage in the development of an Al based image classification model to automate the fragmentation classification process. A camera was installed on the roof of a ramp in an underground sublevel caving mine. The raw images of the truck buckets transporting the ore from underground to surface were captured using a wide angle camera. The raw images were filtered and sorted manually. A subset comprising approximately 1100 sorted images was manually labelled to compile a training data set. The blast-induced fragmented material was classified into five classes: very fine, fine, medium, coarse, and very coarse. The labelled data were used as input to train a deep neural network for image classification. The output of the classification model is the identified material class along with its corresponding confidence score. The developed model was tested on new images from the same mine. The paper also discusses model training on low resolution images, model development and testing, and challenges in adoption and scalability.