Classification of Deep-SAT Images under Label Noise
One of the challenges of training artificial intelligence models for classifying satellite images is the presence of label noise in the datasets that are sometimes crowd-source labeled and as a result, somewhat error prone. In our work, we have utilized three labeled satellite image datasets namely,...
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Taylor & Francis Group
2021-12-01
|
| Series: | Applied Artificial Intelligence |
| Online Access: | http://dx.doi.org/10.1080/08839514.2021.1975381 |
| _version_ | 1797684853938847744 |
|---|---|
| author | Mohammad Minhazul Alam Md Gazuruddin Nahian Ahmed Abdul Motaleb Masud Rana Romman Riyadh Shishir Sabrina Yeasmin Rashedur M. Rahman |
| author_facet | Mohammad Minhazul Alam Md Gazuruddin Nahian Ahmed Abdul Motaleb Masud Rana Romman Riyadh Shishir Sabrina Yeasmin Rashedur M. Rahman |
| author_sort | Mohammad Minhazul Alam |
| collection | DOAJ |
| description | One of the challenges of training artificial intelligence models for classifying satellite images is the presence of label noise in the datasets that are sometimes crowd-source labeled and as a result, somewhat error prone. In our work, we have utilized three labeled satellite image datasets namely, SAT-6, SAT-4, and EuroSAT. The combined dataset consists of over 900,000 image patches that belong to a land cover class. We have applied some standard pixel-based feature extraction algorithms to extract features from the images and then trained those features with various machine learning algorithms. In our experiment, three types of artificial label noises are injected – Noise Completely at Random (NCAR), Noise at Random (NAR) and Noise Not at Random (NNAR) to the training datasets. The noisy data are used to train the algorithms, and the effect of noise on the algorithm performance are compared with noise-free test sets. From our study, the Random Forest and the Back-propagation Neural Network classifiers are found to be the least sensitive to label noises. As label noise is a common scenario in human-labeled image datasets, the current research initiative will help the development of noise robust classification methods for various relevant applications. |
| first_indexed | 2024-03-12T00:35:43Z |
| format | Article |
| id | doaj.art-08dc29f18e2846a4b118526e066af91f |
| institution | Directory Open Access Journal |
| issn | 0883-9514 1087-6545 |
| language | English |
| last_indexed | 2024-03-12T00:35:43Z |
| publishDate | 2021-12-01 |
| publisher | Taylor & Francis Group |
| record_format | Article |
| series | Applied Artificial Intelligence |
| spelling | doaj.art-08dc29f18e2846a4b118526e066af91f2023-09-15T09:33:59ZengTaylor & Francis GroupApplied Artificial Intelligence0883-95141087-65452021-12-0135141196121810.1080/08839514.2021.19753811975381Classification of Deep-SAT Images under Label NoiseMohammad Minhazul Alam0Md Gazuruddin1Nahian Ahmed2Abdul Motaleb3Masud Rana4Romman Riyadh Shishir5Sabrina Yeasmin6Rashedur M. Rahman7North South UniversityNorth South UniversityOregon State UniversityNorth South UniversityNorth South UniversityNorth South UniversityNorth South UniversityNorth South UniversityOne of the challenges of training artificial intelligence models for classifying satellite images is the presence of label noise in the datasets that are sometimes crowd-source labeled and as a result, somewhat error prone. In our work, we have utilized three labeled satellite image datasets namely, SAT-6, SAT-4, and EuroSAT. The combined dataset consists of over 900,000 image patches that belong to a land cover class. We have applied some standard pixel-based feature extraction algorithms to extract features from the images and then trained those features with various machine learning algorithms. In our experiment, three types of artificial label noises are injected – Noise Completely at Random (NCAR), Noise at Random (NAR) and Noise Not at Random (NNAR) to the training datasets. The noisy data are used to train the algorithms, and the effect of noise on the algorithm performance are compared with noise-free test sets. From our study, the Random Forest and the Back-propagation Neural Network classifiers are found to be the least sensitive to label noises. As label noise is a common scenario in human-labeled image datasets, the current research initiative will help the development of noise robust classification methods for various relevant applications.http://dx.doi.org/10.1080/08839514.2021.1975381 |
| spellingShingle | Mohammad Minhazul Alam Md Gazuruddin Nahian Ahmed Abdul Motaleb Masud Rana Romman Riyadh Shishir Sabrina Yeasmin Rashedur M. Rahman Classification of Deep-SAT Images under Label Noise Applied Artificial Intelligence |
| title | Classification of Deep-SAT Images under Label Noise |
| title_full | Classification of Deep-SAT Images under Label Noise |
| title_fullStr | Classification of Deep-SAT Images under Label Noise |
| title_full_unstemmed | Classification of Deep-SAT Images under Label Noise |
| title_short | Classification of Deep-SAT Images under Label Noise |
| title_sort | classification of deep sat images under label noise |
| url | http://dx.doi.org/10.1080/08839514.2021.1975381 |
| work_keys_str_mv | AT mohammadminhazulalam classificationofdeepsatimagesunderlabelnoise AT mdgazuruddin classificationofdeepsatimagesunderlabelnoise AT nahianahmed classificationofdeepsatimagesunderlabelnoise AT abdulmotaleb classificationofdeepsatimagesunderlabelnoise AT masudrana classificationofdeepsatimagesunderlabelnoise AT rommanriyadhshishir classificationofdeepsatimagesunderlabelnoise AT sabrinayeasmin classificationofdeepsatimagesunderlabelnoise AT rashedurmrahman classificationofdeepsatimagesunderlabelnoise |