Objective evaluation-based efficient learning framework for hyperspectral image classification

Deep learning techniques with remarkable performance have been successfully applied to hyperspectral image (HSI) classification. Due the limited availability of training data, earlier studies primarily adopted patch-based classification framework, which divides images into overlapping patches for and testing. However, this framework results in redundant computations possible information leakage. This study proposes an objective evaluation-based efficient HSI It consists two main parts: (i) a leakage-free balanced sampling strategy (ii) fully convolutional network (EfficientFCN) optimized accuracy-efficiency trade-off. The first generates non-overlapping test data by partitioning its ground truth windows. Then, generated are used train proposed EfficientFCN. EfficientFCN exhibits pixel-to-pixel architecture modifications faster inference speed improved parameter efficiency. Experimental demonstrate that can provide evaluation. outperforms many state-of-the-art approaches concerning speed-accuracy For instance, compared recent models EfficientNetV2 ConvNeXt, achieves 0.92% 3.42% superior accuracy 0.19s 0.16s time, respectively, on Houston dataset. Code is available at https://github.com/xmzhang2018.