Hunan Provincial Key Laboratory of Intelligent Processing of Big Data on Transportation, School ofComputer and Communication Engineering, Changsha University of Science and Technology, Changsha 410114, China.
School of Information Science and Engineering, Fujian University of Technology, Fuzhou 350118, China.
Sensors (Basel). 2020 Feb 21;20(4):1188. doi: 10.3390/s20041188.
Many remote sensing scene classification algorithms improve their classification accuracyby additional modules, which increases the parameters and computing overhead of the model atthe inference stage. In this paper, we explore how to improve the classification accuracy of themodel without adding modules at the inference stage. First, we propose a network trainingstrategy of training with multi-size images. Then, we introduce more supervision information bytriplet loss and design a branch for the triplet loss. In addition, dropout is introduced between thefeature extractor and the classifier to avoid over-fitting. These modules only work at the trainingstage and will not bring about the increase in model parameters at the inference stage. We useResnet18 as the baseline and add the three modules to the baseline. We perform experiments onthree datasets: , and . Experimental results show that our modelcombined with the three modules is more competitive than many existing classification algorithms.In addition, ablation experiments on show that dropout, triplet loss, and training withmulti-size images improve the overall accuracy of the model on the test set by 0.53%, 0.38%, and0.7%, respectively. The combination of the three modules improves the overall accuracy of themodel by 1.61%. It can be seen that the three modules can improve the classification accuracy of themodel without increasing model parameters at the inference stage, and training with multi-sizeimages brings a greater gain in accuracy than the other two modules, but the combination of thethree modules will be better.
许多遥感场景分类算法通过附加模块来提高分类精度，这会增加模型在推断阶段的参数和计算开销。在本文中，我们探讨了如何在不增加推断阶段模块的情况下提高模型的分类精度。首先，我们提出了一种使用多尺寸图像进行网络训练的策略。然后，我们通过三元组损失引入更多的监督信息，并为三元组损失设计一个分支。此外，在特征提取器和分类器之间引入了 dropout 以避免过拟合。这些模块仅在训练阶段起作用，不会导致模型参数在推断阶段增加。我们使用 Resnet18 作为基线，并在基线中添加了三个模块。我们在三个数据集上进行了实验：, 和 。实验结果表明，我们的模型结合了这三个模块，比许多现有的分类算法更具竞争力。此外，在 上的消融实验表明，dropout、三元组损失和多尺寸图像训练分别使模型在测试集上的整体准确率提高了 0.53%、0.38%和 0.7%。三个模块的结合使模型的整体准确率提高了 1.61%。可以看出，这三个模块可以在不增加推断阶段模型参数的情况下提高模型的分类精度，并且多尺寸图像训练带来的精度增益大于其他两个模块，但三个模块的结合会更好。
