Blind and visually impaired (BVI) face various problems during their navigation. Being unable to rely on sight greatly restricts their capacity to learn information about their surroundings. Scene recognition is crucial in improving life quality for BVI. Scene recognition systems can recognize and characterize the visual world using powerful artificial intelligence algorithms, allowing users to receive a critical overview. Indoor scene recognition systems are crucial for BVI to explore their environment. These systems are essential for increasing their independence, safety, and overall quality of life. Developing technology that allows BVI to perceive their environment and enable them to navigate and interact with the world on their own is extremely important. We propose in this paper a scene recognition system to assist BVI in their daily activities. The proposed work was developed on top of an efficient set of deep learning techniques called “Deep Evolutionary Algorithms (DEAs)”. DEAs are a type of algorithm that solves complicated search problems by combining the principles of evolutionary computing with deep learning. DEAs provide optimization techniques inspired by the process of natural selection. They iteratively develop a population of potential networks to identify optimum or near-optimal networks to solve complicated problems through genetic methods including mutation, crossover, and selection. To ensure the efficiency of the proposed work, extensive experiments have been conducted using two benchmark datasets the MIT 67 dataset and the Scene 15 dataset. New state-of-the-art results have been ensured by the proposed work in terms of recognition accuracy.