A novel hybrid YOLO-O SegNet for object detection and optimization with DCNN-based object recognition in federated learning

One of the significant Artificial Intelligence (AI) methods for safety monitoring applications is visual object detection. Various techniques have been employed for object detection but they may not work well for indoor safety monitoring images with unique challenges, such as poor lighting, overlapping objects, and unusual object shapes. Therefore, a novel Federated Learning (FL) technique named Fractional Political-Smart Flower Optimization Algorithm Deep Convolutional Neural Network (FP-SFOA_DCNN) with YOLO v3_O-SegNet is introduced for object recognition and detection. At local node, the input indoor image is gained from particular dataset, and it is fed to preprocessing step. The preprocessing is carried out by an Adaptive Bilateral Filter (ABF) and then, object detection is done by proposed YOLO v3_O-SegNet. Feature extraction is performed and then object recognition is done by DCNN, which is trained by FP-SFOA. The FP-SFOA is formed by the combination of Smart Flower Optimization Algorithm (SFOA), Political Optimization (PO), and Fractional Calculus (FC). Finally, local updation and aggregation at the server is performed based on the Conditional Autoregressive Value at Risk (CAViaR). The FP-SFOA-DCNN recorded accuracy, loss, Mean Square Error (MSE), Root Mean Square Error (RMSE), False Positive Rate (FPR), mean average precision, and communication cost of 95.50%, 0.044, 0.109, 0.330, 12.65%, 92.28% and 9.983 respectively.