Abstract: The harsh operating environment inside mineral processing and screening workshops gives rise to problems such as high risks,low efficiency in the detection of vibrating screen discharge outlets,and difficulties in establishing remote monitoring platforms. Therefore,a method combining mechanical devices with computer vision was proposed to automatically detect abnormal states of the outlet. First,a patrol robot platform with a rigid telescopic structure was built to carry and protect cameras while enabling free,flexible,and close-range image sampling. Meanwhile,a rack commutator was designed to reduce the diameter of the telescopic tube,thereby solving the problem of excessive load caused by rigid materials. Second,an image recognition model combining support vector data description(SVDD)with a neural network was constructed. The Le Net network was used as the feature extractor,and an autoencoder was adopted to pre-train LeNet for improving the network convergence speed. Experimental results show that the proposed method achieves an anomaly detection accuracy of 98.7% and an AUC value of 0.99 for discharge outlets,which meets the actual production requirements.