A High-Precision Real-Time Semantic Segmentation Algorithm Based on Big Data and BPNN

The low resolution feature of the original real-time semantic image will interfere with the integrity of the semantic segmentation region, which has the problems of large gap between the segmentation results and the reality, and high average semantic recognition error rate. Therefore, a high-precision realtime semantic segmentation algorithm based on big data and BPNN is proposed. Combining big data analysis and deep learning technology, we can obtain multi-scale spatial information and extract spatial details. BPNN is introduced as the core module of feature fusion to dynamically adjust the weight of low-resolution features and realize the fusion of low-resolution semantic information. The detail branch and semantic branch are designed, and the information is effectively integrated through innovative fusion module and deep separable convolution to achieve highprecision real-time semantic segmentation. The experimental results show that the segmentation results of this algorithm on the dataset are consistent with the real one, the integrity of the segmented region is high, and the average semantic recognition error rate is only 3.17%, which is effective.