Preventing Routing Attacks in Wireless Sensor Networks With Blockchain and Attention‐Based Residual Depth‐Wise Separable Convolution and Banyan Tree Growth Optimization

Wireless sensor networks (WSNs) play a crucial role in healthcare and environmental monitoring, but are prone to attacks such as wormhole, sinkhole, and selective forwarding. Routing attacks cause network performance degradation and violate data integrity. Conventional deep learning models suffer from low accuracy in identifying sophisticated patterns of attacks and have high computational overhead, rendering them inappropriate for WSNs with resource constraints. The energy‐intensive nature of WSN nodes and their scarce computational capacity necessitate more optimal models to support routing security in real time. The solution herein presents a quaternion attention‐based residual depth‐wise separable convolutional network (QARSCN‐BTGO) to address these limitations. This solution utilizes quaternion algebra to speedily process multi‐dimensional sensor data in reduced computational complexity, while boosting feature extraction. Dynamic prioritization of essential routing information is facilitated by an attention mechanism for improved attack detection accuracy. Residual connections avoid loss of information and facilitate smooth gradient flow, while depth‐wise separable convolution decreases parameters, reducing computational burden. For even better performance, banyan tree growth optimization (BTGO) is utilized, which optimizes hyperparameters. BTGO enhances network convergence rate and accuracy while reducing computational time and power consumption. Simulation experiments prove that this method is more accurate and efficient than current models and thus a viable solution for avoiding routing attacks in WSNs.