An Experiment Study on Optimal Batch size of U-Net Convolutional Neural Network for Edge detection

The U-Net architecture, initially designed for biomedical image segmentation, canbe repurposed for edge detection tasks by reconfiguring the network’s focus. This paperinvestigates the impact of batch size, a critical hyperparameter in U-Net, on the network’sperformance for edge detection. We conduct experiments with three different image sizes andvarying batch sizes for each image size. By analyzing the trade-offs between accuracy andstability, we identify the optimal batch size that enhances the U-Net model’s performance inedge detection tasks. Our study contributes valuable insights into effectively configuring batchsize to improve U-Net’s performance in edge detection applications.

کلیدواژه ها:

U-Net ، Conventional neural network ، Edge detection ، Batch size