Abstract: In response to the problem of difficult real-time and accurate detection of temperature rise anomalies inside low-voltage cable branch boxes, a research on temperature rise anomaly detection method for low-voltage cable branch boxes based on distributed fiber optic sensing and lightweight CNN is carried out. Deploy a distributed fiber optic sensing system to achieve continuous temperature monitoring along the path of branch boxes and obtain high spatiotemporal resolution temperature rise data; A lightweight CNN model based on depthwise separable convolution is used to adaptively extract multidimensional spatiotemporal features that characterize temperature rise anomalies; On this basis, an end-to-end intelligent diagnostic framework was constructed from raw sensor data to fault state discrimination, achieving online intelligent recognition and classification of abnormal temperature rise patterns. Comparative experiments show that this method significantly improves the accuracy of temperature rise anomaly detection and has strong real-time processing efficiency, providing an effective solution for early fault warning of key equipment in distribution networks.