Impact of Dataset Characteristics on the Accuracy of SAR Image Target Detection

The quality of a dataset significantly impacts the performance of algorithm models, yet a unified standard for SAR image dataset production remains absent. Currently, there is no consensus on the optimal image type, bit depth, and preprocessing methods to maximize algorithm effectiveness. This paper explores how these factors—image type, bit depth, and preprocessing techniques—affect target detection models by creating and training multiple SAR image datasets. The results reveal that 8-bit Tagged Image File Forma (TIFF) images, enhanced with linear stretching of the maximum and minimum values, achieve the best performance. These findings offer a valuable reference for establishing standards in the production of future SAR image datasets.