Performance Analysis of CPUs and GPUs for Training CNN models using Tomato Leaf Diseases

Plant diseases are a severe threat to plant development and food security, making early detection is crucial for preventive actions. Convolutional Neural Networks (CNNs) are considered a promising tool for the automated detection of plant diseases. However, the high computational demands associated with CNN training pose significant challenges, necessitating the importance of selecting hardware platforms carefully for optimal performance and cost-effectiveness. The study evaluates CNN performance on Central Processing Unit (CPU) and Graphics Processing Unit (GPU) platforms for plant disease detection. Key performance metrics including training time, accuracy, memory usage, and throughput are compared between CPU and GPU implementations to evaluate their respective efficiencies. Insights gained from this outcome is that, GPU demonstrate superior performance over CPU by achieving significantly faster training time of 86.27seconds and higher throughput of 87.88 samples/second despite operating at a lower clock speed of 0.000585GHZ. The study underscores the advantage of GPU acceleration for CNN based plant disease detection, offering valuable insights for optimizing real-world implementation.