Belt conveyor systems are essential in industrial bulk material transport due to their high capacity and automation. However, foreign objects on the belt can disrupt operations and damage downstream equipment. Manual inspection is limited, highlighting the need for continuous automated monitoring. This paper presents a multi-step solution using video stream data from a camera above the conveyor. Frames are pre-processed to detect belt activity and define a region of interest (ROI) based on ore edges. An object detection pipeline, enhanced with image processing techniques, identifies foreign objects. Inconsistent edge detection can cause false positives, especially due to dark thresholding. To address this, a convolutional neural network (CNN) using MobileNetV2 pre-trained weights refines detection accuracy. Deployed at LKAB Narvik, the system integrates with a SaaS platform for predictive maintenance and decision support. It significantly reduces false detections and effectively identifies both high- and low-intensity objects, improving operational reliability.