The invention discloses a computer-vision-based self-adaptive control method for a belt conveyor for a coal wharf. The computer-vision-based self-adaptive control method for the belt conveyor for thecoal wharf comprises the steps of during the off-line pre-training stage, through recording a running video of a conveyor belt, extracting an image, making a conveying capacity training data set, a conveying capacity test data set, an abnormal state training data set and an abnormal state testing data set, and training a belt conveyor conveying capacity detection model and a belt conveyor abnormalstate detection model based on a convolutional neural network; during the on-line monitoring stage, using the trained belt conveyor conveying capacity detection model and the trained belt conveyor abnormal state detection model for monitoring the conveyor belt in real time, self-adaptively controlling the conveying speed of the belt conveyor according to the conveying capacity of the belt conveyor, detecting multiple abnormal states of the conveyor belt at the same time, immediately stopping running after the abnormal states are found, and sending an alarm corresponding to the abnormal state.The operation cost of the belt conveyor is greatly reduced, the overhaul and maintenance efficiency of the belt conveyor are improved, the maintenance cost is reduced, the safety is improved, the detection models are built by using the convolutional neural network, the detection accuracy is improved, and the recognition error rate is reduced.