Remote Sensing Scene Classification and Explanation Using RSSCNet and LIME

Abstract

Classification is needed in disaster investigation, traffic control, and land-use resource management. How to quickly and accurately classify such remote sensing imagery has become a popular research topic. However, the application of large, deep neural network models for the training of classifiers in the hope of obtaining good classification results is often very time-consuming. In this study, a new CNN (convolutional neutral networks) architecture, i.e., RSSCNet (remote sensing scene classification network), with high generalization capability was designed. Moreover, a two-stage cyclical learning rate policy and the no-freezing transfer learning method were developed to speed up model training and enhance accuracy. In addition, the manifold learning t-SNE (t-distributed stochastic neighbor embedding) algorithm was used to verify the effectiveness of the proposed model, and the LIME (local interpretable model, agnostic explanation) algorithm was applied to improve the results in cases where the model made wrong predictions. Comparing the results of three publicly available datasets in this study with those obtained in previous studies, the experimental results show that the model and method proposed in this paper can achieve better scene classification more quickly and more efficiently.

A Deep Learning Model for Classification of Endoscopic Gastroesophageal Reflux Disease

“…In consideration of the balance between network capacity and validation accuracy and by discussing the influence of different regularization and optimization strategies [ 26 ], we designed a deep convolutional neural network (CNN) architecture with high generalizability. The proposed CNN architecture is called GERD-VGGNet ( Figure 3 ).…”

“…The proposed GERD-VGGNet architecture includes 13 convolutional layers, five max pooling layers, one global average pooling layer, four dense layers, four batch normalization layers [ 27 ], four activation layers using the rectified linear unit (ReLU) function [ 26 , 28 ], and the last dense layer with softmax classification.…”

“…This model employs the Adam optimizer with a batch size of 64 examples for two-stage optimization training of the entire network architecture using a non-freezing transfer learning method [ 26 ]. Here, in the first training stage, the learning rate is set to for network training up to 600 epochs.…”

“…To enhance the generalizability of the model, dynamic data augmentation is considered an effective method to train a generally applicable model using a limited amount of training data [ 26 , 29 ]. To make up for the lack of data, a dynamic data augmentation technique is included in the training.…”

RS-SSKD: Self-Supervision Equipped with Knowledge Distillation for Few-Shot Remote Sensing Scene Classification

“…Despite these and other state-of-the-art methods have made significant progress to date and have even achieved ∼100% accuracy on some datasets (e.g., [ 24 ] achieves 99.82% accuracy on the UC Merced dataset [ 16 , 31 ] achieves 99.46% on the WHU-RS19 dataset [ 27 ]), one may argue, is that machine learning really outperforming human performance, or is the dataset too simple? For example, the UC Merced dataset holds 21 scene classes with 100 images per class, while the WHU-RS19 dataset contains 19 classes with ∼50 images in each.…”