The Analysis of Industrial Heritage Landscape Design in Tourism Based on Virtual Reality and Convolutional Neural Network

To enhance the efficiency and performance of virtual reality systems applied to industrial heritage landscape design, this study introduces an optimization model based on MobileNetV4, accompanied by a thorough experimental evaluation. A series of comparative experiments reveals that this model excels in various critical metrics, including model accuracy, inference speed, model size, and computational complexity. For a dataset comprising 5,000 samples, the model achieves an impressive accuracy of 0.910, an inference speed of 30.787 ms, a compact model size of 5.216 MB, and a computational complexity of 0.177 GFLOPS. Additionally, the contributions of pivotal components to the model's performance were meticulously assessed through ablation studies. Results indicate that the elimination of the depth-separable convolution layer resulted in a slowdown of inference speed to 33.090 ms and an increase in computational complexity to 0.204 GFLOPS.