A Remote-Sensing Scene-Image Classification Method Based on Deep Multiple-Instance Learning with a Residual Dense Attention ConvNet
The spatial distribution of remote-sensing scene images is highly complex in character, so how to extract local key semantic information and discriminative features is the key to making it possible to classify accurately. However, most of the existing convolutional neural network (CNN) models tend t...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2022-10-01
|
| Series: | Remote Sensing |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2072-4292/14/20/5095 |
| _version_ | 1797470106828144640 |
|---|---|
| author | Xinyu Wang Haixia Xu Liming Yuan Wei Dai Xianbin Wen |
| author_facet | Xinyu Wang Haixia Xu Liming Yuan Wei Dai Xianbin Wen |
| author_sort | Xinyu Wang |
| collection | DOAJ |
| description | The spatial distribution of remote-sensing scene images is highly complex in character, so how to extract local key semantic information and discriminative features is the key to making it possible to classify accurately. However, most of the existing convolutional neural network (CNN) models tend to have global feature representations and lose the shallow features. In addition, when the network is too deep, gradient disappearance and overfitting tend to occur. To solve these problems, a lightweight, multi-instance CNN model for remote sensing scene classification is proposed in this paper: MILRDA. In the instance extraction and classifier part, more discriminative features are extracted by the constructed residual dense attention block (RDAB) while retaining shallow features. Then, the extracted features are transformed into instance-level vectors and the local information associated with bag-level labels is highlighted by the proposed channel-attention-based multi-instance pooling, while suppressing the weights of useless objects or backgrounds. Finally, the network is constrained by the cross-entropy loss function to output the final prediction results. The experimental results on four public datasets show that our proposed method can achieve comparable results to other state-of-the-art methods. Moreover, the visualization of feature maps shows that MILRDA can find more effective features. |
| first_indexed | 2024-03-09T19:32:03Z |
| format | Article |
| id | doaj.art-c312f61102df4a408b9f975ead9270c3 |
| institution | Directory Open Access Journal |
| issn | 2072-4292 |
| language | English |
| last_indexed | 2024-03-09T19:32:03Z |
| publishDate | 2022-10-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Remote Sensing |
| spelling | doaj.art-c312f61102df4a408b9f975ead9270c32023-11-24T02:19:06ZengMDPI AGRemote Sensing2072-42922022-10-011420509510.3390/rs14205095A Remote-Sensing Scene-Image Classification Method Based on Deep Multiple-Instance Learning with a Residual Dense Attention ConvNetXinyu Wang0Haixia Xu1Liming Yuan2Wei Dai3Xianbin Wen4School of Computer Science and Engineering, and Key Laboratory of Computer Vision, Tianjin University of Technology, Tianjin 300384, ChinaSchool of Computer Science and Engineering, and Key Laboratory of Computer Vision, Tianjin University of Technology, Tianjin 300384, ChinaSchool of Computer Science and Engineering, and Key Laboratory of Computer Vision, Tianjin University of Technology, Tianjin 300384, ChinaSchool of Computer Science and Engineering, and Key Laboratory of Computer Vision, Tianjin University of Technology, Tianjin 300384, ChinaSchool of Computer Science and Engineering, and Key Laboratory of Computer Vision, Tianjin University of Technology, Tianjin 300384, ChinaThe spatial distribution of remote-sensing scene images is highly complex in character, so how to extract local key semantic information and discriminative features is the key to making it possible to classify accurately. However, most of the existing convolutional neural network (CNN) models tend to have global feature representations and lose the shallow features. In addition, when the network is too deep, gradient disappearance and overfitting tend to occur. To solve these problems, a lightweight, multi-instance CNN model for remote sensing scene classification is proposed in this paper: MILRDA. In the instance extraction and classifier part, more discriminative features are extracted by the constructed residual dense attention block (RDAB) while retaining shallow features. Then, the extracted features are transformed into instance-level vectors and the local information associated with bag-level labels is highlighted by the proposed channel-attention-based multi-instance pooling, while suppressing the weights of useless objects or backgrounds. Finally, the network is constrained by the cross-entropy loss function to output the final prediction results. The experimental results on four public datasets show that our proposed method can achieve comparable results to other state-of-the-art methods. Moreover, the visualization of feature maps shows that MILRDA can find more effective features.https://www.mdpi.com/2072-4292/14/20/5095remote-sensing scene image classificationconvolutional neural network (CNN)multiple instance learning (MIL)attention mechanisms |
| spellingShingle | Xinyu Wang Haixia Xu Liming Yuan Wei Dai Xianbin Wen A Remote-Sensing Scene-Image Classification Method Based on Deep Multiple-Instance Learning with a Residual Dense Attention ConvNet Remote Sensing remote-sensing scene image classification convolutional neural network (CNN) multiple instance learning (MIL) attention mechanisms |
| title | A Remote-Sensing Scene-Image Classification Method Based on Deep Multiple-Instance Learning with a Residual Dense Attention ConvNet |
| title_full | A Remote-Sensing Scene-Image Classification Method Based on Deep Multiple-Instance Learning with a Residual Dense Attention ConvNet |
| title_fullStr | A Remote-Sensing Scene-Image Classification Method Based on Deep Multiple-Instance Learning with a Residual Dense Attention ConvNet |
| title_full_unstemmed | A Remote-Sensing Scene-Image Classification Method Based on Deep Multiple-Instance Learning with a Residual Dense Attention ConvNet |
| title_short | A Remote-Sensing Scene-Image Classification Method Based on Deep Multiple-Instance Learning with a Residual Dense Attention ConvNet |
| title_sort | remote sensing scene image classification method based on deep multiple instance learning with a residual dense attention convnet |
| topic | remote-sensing scene image classification convolutional neural network (CNN) multiple instance learning (MIL) attention mechanisms |
| url | https://www.mdpi.com/2072-4292/14/20/5095 |
| work_keys_str_mv | AT xinyuwang aremotesensingsceneimageclassificationmethodbasedondeepmultipleinstancelearningwitharesidualdenseattentionconvnet AT haixiaxu aremotesensingsceneimageclassificationmethodbasedondeepmultipleinstancelearningwitharesidualdenseattentionconvnet AT limingyuan aremotesensingsceneimageclassificationmethodbasedondeepmultipleinstancelearningwitharesidualdenseattentionconvnet AT weidai aremotesensingsceneimageclassificationmethodbasedondeepmultipleinstancelearningwitharesidualdenseattentionconvnet AT xianbinwen aremotesensingsceneimageclassificationmethodbasedondeepmultipleinstancelearningwitharesidualdenseattentionconvnet AT xinyuwang remotesensingsceneimageclassificationmethodbasedondeepmultipleinstancelearningwitharesidualdenseattentionconvnet AT haixiaxu remotesensingsceneimageclassificationmethodbasedondeepmultipleinstancelearningwitharesidualdenseattentionconvnet AT limingyuan remotesensingsceneimageclassificationmethodbasedondeepmultipleinstancelearningwitharesidualdenseattentionconvnet AT weidai remotesensingsceneimageclassificationmethodbasedondeepmultipleinstancelearningwitharesidualdenseattentionconvnet AT xianbinwen remotesensingsceneimageclassificationmethodbasedondeepmultipleinstancelearningwitharesidualdenseattentionconvnet |