1
|
Hyperspectral Image Classification Based on Class-Incremental Learning with Knowledge Distillation. REMOTE SENSING 2022. [DOI: 10.3390/rs14112556] [Citation(s) in RCA: 1] [Impact Index Per Article: 0.5] [Reference Citation Analysis] [Abstract] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 02/04/2023]
Abstract
By virtue of its large-covered spatial information and high-resolution spectral information, hyperspectral images make lots of mapping-based fine-grained remote sensing applications possible. However, due to the inconsistency of land-cover types between different images, most hyperspectral image classification methods keep their effectiveness by training on every image and saving all classification models and training samples, which limits the promotion of related remote sensing tasks. To deal with the aforementioned issues, this paper proposes a hyperspectral image classification method based on class-incremental learning to learn new land-cover types without forgetting the old ones, which enables the classification method to classify all land-cover types with one final model. Specially, when learning new classes, a knowledge distillation strategy is designed to recall the information of old classes by transferring knowledge to the newly trained network, and a linear correction layer is proposed to relax the heavy bias towards the new class by reapportioning information between different classes. Additionally, the proposed method introduces a channel attention mechanism to effectively utilize spatial–spectral information by a recalibration strategy. Experimental results on the three widely used hyperspectral images demonstrate that the proposed method can identify both new and old land-cover types with high accuracy, which proves the proposed method is more practical in large-coverage remote sensing tasks.
Collapse
|
2
|
Hyperspectral Image Classification via a Novel Spectral–Spatial 3D ConvLSTM-CNN. REMOTE SENSING 2021. [DOI: 10.3390/rs13214348] [Citation(s) in RCA: 3] [Impact Index Per Article: 1.0] [Reference Citation Analysis] [Abstract] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 12/11/2022]
Abstract
In recent years, deep learning-based models have produced encouraging results for hyperspectral image (HSI) classification. Specifically, Convolutional Long Short-Term Memory (ConvLSTM) has shown good performance for learning valuable features and modeling long-term dependencies in spectral data. However, it is less effective for learning spatial features, which is an integral part of hyperspectral images. Alternatively, convolutional neural networks (CNNs) can learn spatial features, but they possess limitations in handling long-term dependencies due to the local feature extraction in these networks. Considering these factors, this paper proposes an end-to-end Spectral-Spatial 3D ConvLSTM-CNN based Residual Network (SSCRN), which combines 3D ConvLSTM and 3D CNN for handling both spectral and spatial information, respectively. The contribution of the proposed network is twofold. Firstly, it addresses the long-term dependencies of spectral dimension using 3D ConvLSTM to capture the information related to various ground materials effectively. Secondly, it learns the discriminative spatial features using 3D CNN by employing the concept of the residual blocks to accelerate the training process and alleviate the overfitting. In addition, SSCRN uses batch normalization and dropout to regularize the network for smooth learning. The proposed framework is evaluated on three benchmark datasets widely used by the research community. The results confirm that SSCRN outperforms state-of-the-art methods with an overall accuracy of 99.17%, 99.67%, and 99.31% over Indian Pines, Salinas, and Pavia University datasets, respectively. Moreover, it is worth mentioning that these excellent results were achieved with comparatively fewer epochs, which also confirms the fast learning capabilities of the SSCRN.
Collapse
|
3
|
Marchewka A, Ziółkowski P, Aguilar-Vidal V. Framework for Structural Health Monitoring of Steel Bridges by Computer Vision. SENSORS 2020; 20:s20030700. [PMID: 32012791 PMCID: PMC7039231 DOI: 10.3390/s20030700] [Citation(s) in RCA: 14] [Impact Index Per Article: 3.5] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Subscribe] [Scholar Register] [Received: 12/24/2019] [Revised: 01/12/2020] [Accepted: 01/21/2020] [Indexed: 11/16/2022]
Abstract
The monitoring of a structural condition of steel bridges is an important issue. Good condition of infrastructure facilities ensures the safety and economic well-being of society. At the same time, due to the continuous development, rising wealth of the society and socio-economic integration of countries, the number of infrastructural objects is growing. Therefore, there is a need to introduce an easy-to-use and relatively low-cost method of bridge diagnostics. We can achieve these benefits by the use of Unmanned Aerial Vehicle-Based Remote Sensing and Digital Image Processing. In our study, we present a state-of-the-art framework for Structural Health Monitoring of steel bridges that involves literature review on steel bridges health monitoring, drone route planning, image acquisition, identification of visual markers that may indicate a poor condition of the structure and determining the scope of applicability. The presented framework of image processing procedure is suitable for diagnostics of steel truss riveted bridges. In our considerations, we used photographic documentation of the Fitzpatrick Bridge located in Tallassee, Alabama, USA.
Collapse
Affiliation(s)
- Adam Marchewka
- Computer Science and Electrical Engineering, Faculty of Telecommunications, University of Science and Technology in Bydgoszcz, Al. prof. S. Kaliskiego 7, 85-796 Bydgoszcz, Poland;
| | - Patryk Ziółkowski
- Faculty of Civil and Environmental Engineering, Gdansk University of Technology, Gabriela Narutowicza 11/12, 80-233 Gdansk, Poland
- Correspondence: ; Tel.: +48-58-347-2385
| | - Victor Aguilar-Vidal
- Department of Civil Engineering, Auburn University, 261 W Magnolia Ave, Auburn, AL 36849, USA;
- Facultad de Ingeniería y Tecnología, Universidad San Sebastián, Lientur 1457, Concepción 4080871, Chile
| |
Collapse
|
4
|
Priego B, Duro RJ. An Approach for the Customized High-Dimensional Segmentation of Remote Sensing Hyperspectral Images. SENSORS 2019; 19:s19132887. [PMID: 31261901 PMCID: PMC6650901 DOI: 10.3390/s19132887] [Citation(s) in RCA: 3] [Impact Index Per Article: 0.6] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Subscribe] [Scholar Register] [Received: 04/02/2019] [Revised: 06/10/2019] [Accepted: 06/23/2019] [Indexed: 11/16/2022]
Abstract
This paper addresses three problems in the field of hyperspectral image segmentation: the fact that the way an image must be segmented is related to what the user requires and the application; the lack and cost of appropriately labeled reference images; and, finally, the information loss problem that arises in many algorithms when high dimensional images are projected onto lower dimensional spaces before starting the segmentation process. To address these issues, the Multi-Gradient based Cellular Automaton (MGCA) structure is proposed to segment multidimensional images without projecting them to lower dimensional spaces. The MGCA structure is coupled with an evolutionary algorithm (ECAS-II) in order to produce the transition rule sets required by MGCA segmenters. These sets are customized to specific segmentation needs as a function of a set of low dimensional training images in which the user expresses his segmentation requirements. Constructing high dimensional image segmenters from low dimensional training sets alleviates the problem of lack of labeled training images. These can be generated online based on a parametrization of the desired segmentation extracted from a set of examples. The strategy has been tested in experiments carried out using synthetic and real hyperspectral images, and it has been compared to state-of-the-art segmentation approaches over benchmark images in the area of remote sensing hyperspectral imaging.
Collapse
Affiliation(s)
- Blanca Priego
- Biomedical Engineering and Telemedicine Researching Group, University of Cádiz, 11002 Cádiz, Spain
- Institute of Research and Innovation in Biomedical Sciences of the Province of Cádiz (INiBICA), University of Cádiz, 11002 Cádiz, Spain
| | - Richard J Duro
- Integrated Group for Engineering Research, Centro de Investigación en Tecnologías de la Información y las Comunicaciones (CITIC), Universidade da Coruña, 15403 Ferrol, Spain.
| |
Collapse
|
5
|
Hyperspectral Image Classification for Land Cover Based on an Improved Interval Type-II Fuzzy C-Means Approach. SENSORS 2018; 18:s18020363. [PMID: 29373548 PMCID: PMC5856092 DOI: 10.3390/s18020363] [Citation(s) in RCA: 17] [Impact Index Per Article: 2.8] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Subscribe] [Scholar Register] [Received: 11/09/2017] [Revised: 01/15/2018] [Accepted: 01/24/2018] [Indexed: 11/17/2022]
Abstract
Few studies have examined hyperspectral remote-sensing image classification with type-II fuzzy sets. This paper addresses image classification based on a hyperspectral remote-sensing technique using an improved interval type-II fuzzy c-means (IT2FCM*) approach. In this study, in contrast to other traditional fuzzy c-means-based approaches, the IT2FCM* algorithm considers the ranking of interval numbers and the spectral uncertainty. The classification results based on a hyperspectral dataset using the FCM, IT2FCM, and the proposed improved IT2FCM* algorithms show that the IT2FCM* method plays the best performance according to the clustering accuracy. In this paper, in order to validate and demonstrate the separability of the IT2FCM*, four type-I fuzzy validity indexes are employed, and a comparative analysis of these fuzzy validity indexes also applied in FCM and IT2FCM methods are made. These four indexes are also applied into different spatial and spectral resolution datasets to analyze the effects of spectral and spatial scaling factors on the separability of FCM, IT2FCM, and IT2FCM* methods. The results of these validity indexes from the hyperspectral datasets show that the improved IT2FCM* algorithm have the best values among these three algorithms in general. The results demonstrate that the IT2FCM* exhibits good performance in hyperspectral remote-sensing image classification because of its ability to handle hyperspectral uncertainty.
Collapse
|
6
|
Coupled compressed sensing inspired sparse spatial-spectral LSSVM for hyperspectral image classification. Knowl Based Syst 2015. [DOI: 10.1016/j.knosys.2015.01.006] [Citation(s) in RCA: 33] [Impact Index Per Article: 3.7] [Reference Citation Analysis] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/20/2022]
|
7
|
Song A, Chang A, Choi J, Choi S, Kim Y. Automatic extraction of optimal endmembers from airborne hyperspectral imagery using iterative error analysis (IEA) and spectral discrimination measurements. SENSORS 2015; 15:2593-613. [PMID: 25625907 PMCID: PMC4367322 DOI: 10.3390/s150202593] [Citation(s) in RCA: 4] [Impact Index Per Article: 0.4] [Reference Citation Analysis] [Abstract] [Track Full Text] [Download PDF] [Figures] [Subscribe] [Scholar Register] [Received: 08/19/2014] [Accepted: 01/19/2015] [Indexed: 11/23/2022]
Abstract
Pure surface materials denoted by endmembers play an important role in hyperspectral processing in various fields. Many endmember extraction algorithms (EEAs) have been proposed to find appropriate endmember sets. Most studies involving the automatic extraction of appropriate endmembers without a priori information have focused on N-FINDR. Although there are many different versions of N-FINDR algorithms, computational complexity issues still remain and these algorithms cannot consider the case where spectrally mixed materials are extracted as final endmembers. A sequential endmember extraction-based algorithm may be more effective when the number of endmembers to be extracted is unknown. In this study, we propose a simple but accurate method to automatically determine the optimal endmembers using such a method. The proposed method consists of three steps for determining the proper number of endmembers and for removing endmembers that are repeated or contain mixed signatures using the Root Mean Square Error (RMSE) images obtained from Iterative Error Analysis (IEA) and spectral discrimination measurements. A synthetic hyperpsectral image and two different airborne images such as Airborne Imaging Spectrometer for Application (AISA) and Compact Airborne Spectrographic Imager (CASI) data were tested using the proposed method, and our experimental results indicate that the final endmember set contained all of the distinct signatures without redundant endmembers and errors from mixed materials.
Collapse
Affiliation(s)
- Ahram Song
- Department of Civil and Environmental Engineering, Seoul National University, 1, Gwanak-ro, Gwanak-gu, Seoul 151-742, Korea.
| | - Anjin Chang
- School of Earth and Environmental Sciences, Seoul National University, 1, Gwanak-ro, Gwanak-gu, Seoul 151-742, Korea.
| | - Jaewan Choi
- School of Civil Engineering, Chungbuk National University, 1 Chungdae-ro , Seowon-gu, Cheongju, Chungbuk 361-763, Korea.
| | - Seokkeun Choi
- School of Civil Engineering, Chungbuk National University, 1 Chungdae-ro , Seowon-gu, Cheongju, Chungbuk 361-763, Korea.
| | - Yongil Kim
- Department of Civil and Environmental Engineering, Seoul National University, 1, Gwanak-ro, Gwanak-gu, Seoul 151-742, Korea.
| |
Collapse
|
8
|
Extended averaged learning subspace method for hyperspectral data classification. SENSORS 2009; 9:4247-70. [PMID: 22408524 PMCID: PMC3291909 DOI: 10.3390/s90604247] [Citation(s) in RCA: 9] [Impact Index Per Article: 0.6] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Subscribe] [Scholar Register] [Received: 05/14/2009] [Revised: 05/27/2009] [Accepted: 06/01/2009] [Indexed: 11/24/2022]
Abstract
Averaged learning subspace methods (ALSM) have the advantage of being easily implemented and appear to outperform in classification problems of hyperspectral images. However, there remain some open and challenging problems, which if addressed, could further improve their performance in terms of classification accuracy. We carried out experiments mainly by using two kinds of improved subspace methods (namely, dynamic and fixed subspace methods), in conjunction with the [0,1] and [-1,+1] normalization methods. We used different performance indicators to support our experimental studies: classification accuracy, computation time, and the stability of the parameter settings. Results are presented for the AVIRIS Indian Pines data set. Experimental analysis showed that the fixed subspace method combined with the [0,1] normalization method yielded higher classification accuracy than other subspace methods. Moreover, ALSMs are easily applied: only two parameters need to be set, and they can be applied directly to hyperspectral data. In addition, they can completely identify training samples in a finite number of iterations.
Collapse
|