Abstract: With the rapid development of the Internet of Things and deep learning, intelligent production operations have become the trend of the new generation of industrial development. The blockage of cargo transshipment in dry bulk ports will significantly reduce the efficiency of transshipment operations. To this end, in this paper, based on the signal data collected by millimeter-wave radar, a semi-supervised anomaly detection model is constructed to detect and identify cargo blockage in the transfer tower. On the premise of ensuring that the data features of the blockage signal are not lost, the model first extracts a high resolution range profile from the collected original millimeter-wave radar signal, and then extracts features by adding a self-attention module and contrastive learning. Finally, the detection model can more accurately distinguish the difference between normal samples and blocked samples. Experimental comparison with multiple related models shows that the detection model proposed in this paper has the best effect on the corresponding task, and can achieve a detection rate of nearly 90% for normal samples.