Seg_UResNet: A Deep Hybrid Convolutional Neural Network of Road Scenes for semantic segmentation

Comprehending the circumstances is an essential element of every autonomous driving system. Computer vision harbors the task of semantic segmentation as one of its core components wherein each pixel in an image is labeled with precise labels. The task has found significant applications in areas like autonomous driving plus analysis of images and videos, not to mention medical imaging and scene understanding. The progress made in semantic segmentation through the utilization of deep learning techniques has been truly extraordinary. Our study introduces a groundbreaking approach known as hybrid evolutionary weighted ensemble, which combines the strengths of two deep models. Two distinct segmentation models will be developed using U-Net and ResNet34 as the decoder and the encoder respectively for the first model using transfer learning technique, we are able to create individual segmentation models. Additionally, we consider SegNet another highly regarded convolutional neural network for image segmentation as our second deep model. With the goal of enhancing the accuracy of the deep learning model, we propose an evolutionary leap towards the convergence of the DL model accuracy will be achieved through the introduction of a hybrid weighted ensemble model. This shall be done by combining ResNet34, an individual U-Net-based DTL model, with SegNet model and then executing fine-tuning. To assess the effectiveness of the hybrid network, we conducted an evaluation using the widely recognized and demanding CamVid dataset. Through our experiments, we were able to showcase that the hybrid network outperforms other popular semantic segmentation algorithms by a significant margin, solidifying its position as a cutting-edge solution.