Improving Multi‐Domain Autonomous Driving Data Classification Using Federated Learning‐Based Yolo With Adaptive Dilated Residual Attention Network

ABSTRACT Usually, autonomous driving is considered to be one of the quickest‐developing, recently, cutting‐edge research domains, with huge market benefits in upcoming days. However, a few drawbacks occur because of the variety and size of the vehicle conditions that can be observed from the realistic world along with a limited number of cars traveling around. In this proposed method, a new adaptive model is introduced for the Multi‐domain autonomous driving data classification which is more helpful for ensuring safety in autonomous driving and also guarantees driving control. The necessary input images are garnered from standard assets. Further, the input images are processed by the Retinex network for managing a climate and light circumstances. The class imbalance of the images is handled by performing the image augmentation process using the attention‐based Generative Adversarial Network (GAN). Finally, the object detection and classification task is executed by Federated Learning‐based Yolo v9 with an “Adaptive Dilated Residual Attention Network (Yolo‐ADRAN).” The detection and classification pipeline is adapted to the adaptive network where a parameter of this method is tune by using a Randomly‐emphasized Swarm Space Hopping Algorithm (RSSHA), thus it rapidly detects and classifies the objects as of the inputs. Lastly, a presentation of the method is confirmed using the experimental analysis.