Constraint Loss for Rotated Object Detection in Remote Sensing Images
Rotated object detection is an extension of object detection that uses an oriented bounding box instead of a general horizontal bounding box to define the object position. It is widely used in remote sensing images, scene text, and license plate recognition. The existing rotated object detection met...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2021-10-01
|
| Series: | Remote Sensing |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2072-4292/13/21/4291 |
| _version_ | 1797511873635024896 |
|---|---|
| author | Luyang Zhang Haitao Wang Lingfeng Wang Chunhong Pan Qiang Liu Xinyao Wang |
| author_facet | Luyang Zhang Haitao Wang Lingfeng Wang Chunhong Pan Qiang Liu Xinyao Wang |
| author_sort | Luyang Zhang |
| collection | DOAJ |
| description | Rotated object detection is an extension of object detection that uses an oriented bounding box instead of a general horizontal bounding box to define the object position. It is widely used in remote sensing images, scene text, and license plate recognition. The existing rotated object detection methods usually add an angle prediction channel in the bounding box prediction branch, and smooth <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><msub><mi>L</mi><mn>1</mn></msub></semantics></math></inline-formula> loss is used as the regression loss function. However, we argue that smooth <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><msub><mi>L</mi><mn>1</mn></msub></semantics></math></inline-formula> loss causes a sudden change in loss and slow convergence due to the angle solving mechanism of open CV (the angle between the horizontal line and the first side of the bounding box in the counter-clockwise direction is defined as the rotation angle), and this problem exists in most existing regression loss functions. To solve the above problems, we propose a decoupling modulation mechanism to overcome the problem of sudden changes in loss. On this basis, we also proposed a constraint mechanism, the purpose of which is to accelerate the convergence of the network and ensure optimization toward the ideal direction. In addition, the proposed decoupling modulation mechanism and constraint mechanism can be integrated into the popular regression loss function individually or together, which further improves the performance of the model and makes the model converge faster. The experimental results show that our method achieves 75.2% performance on the aerial image dataset DOTA (OBB task), and saves more than 30% of computing resources. The method also achieves a state-of-the-art performance in HRSC2016, and saved more than 40% of computing resources, which confirms the applicability of the approach. |
| first_indexed | 2024-03-10T05:54:06Z |
| format | Article |
| id | doaj.art-acd2c41bbb26464b8615e0d8c1436ce8 |
| institution | Directory Open Access Journal |
| issn | 2072-4292 |
| language | English |
| last_indexed | 2024-03-10T05:54:06Z |
| publishDate | 2021-10-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Remote Sensing |
| spelling | doaj.art-acd2c41bbb26464b8615e0d8c1436ce82023-11-22T21:31:17ZengMDPI AGRemote Sensing2072-42922021-10-011321429110.3390/rs13214291Constraint Loss for Rotated Object Detection in Remote Sensing ImagesLuyang Zhang0Haitao Wang1Lingfeng Wang2Chunhong Pan3Qiang Liu4Xinyao Wang5College of Automation Engineering, Nanjing University of Aeronautics & Astronautics, Nanjing 210016, ChinaCollege of Automation Engineering, Nanjing University of Aeronautics & Astronautics, Nanjing 210016, ChinaCollege of Information Science and Technology, Beijing University of Chemical Technology, Beijing 100029, ChinaNational Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing 100190, ChinaCollege of Automation Engineering, Nanjing University of Aeronautics & Astronautics, Nanjing 210016, ChinaCollege of Automation Engineering, Nanjing University of Aeronautics & Astronautics, Nanjing 210016, ChinaRotated object detection is an extension of object detection that uses an oriented bounding box instead of a general horizontal bounding box to define the object position. It is widely used in remote sensing images, scene text, and license plate recognition. The existing rotated object detection methods usually add an angle prediction channel in the bounding box prediction branch, and smooth <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><msub><mi>L</mi><mn>1</mn></msub></semantics></math></inline-formula> loss is used as the regression loss function. However, we argue that smooth <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><msub><mi>L</mi><mn>1</mn></msub></semantics></math></inline-formula> loss causes a sudden change in loss and slow convergence due to the angle solving mechanism of open CV (the angle between the horizontal line and the first side of the bounding box in the counter-clockwise direction is defined as the rotation angle), and this problem exists in most existing regression loss functions. To solve the above problems, we propose a decoupling modulation mechanism to overcome the problem of sudden changes in loss. On this basis, we also proposed a constraint mechanism, the purpose of which is to accelerate the convergence of the network and ensure optimization toward the ideal direction. In addition, the proposed decoupling modulation mechanism and constraint mechanism can be integrated into the popular regression loss function individually or together, which further improves the performance of the model and makes the model converge faster. The experimental results show that our method achieves 75.2% performance on the aerial image dataset DOTA (OBB task), and saves more than 30% of computing resources. The method also achieves a state-of-the-art performance in HRSC2016, and saved more than 40% of computing resources, which confirms the applicability of the approach.https://www.mdpi.com/2072-4292/13/21/4291rotated object detectionremote sensing imageloss functionsfast convergence |
| spellingShingle | Luyang Zhang Haitao Wang Lingfeng Wang Chunhong Pan Qiang Liu Xinyao Wang Constraint Loss for Rotated Object Detection in Remote Sensing Images Remote Sensing rotated object detection remote sensing image loss functions fast convergence |
| title | Constraint Loss for Rotated Object Detection in Remote Sensing Images |
| title_full | Constraint Loss for Rotated Object Detection in Remote Sensing Images |
| title_fullStr | Constraint Loss for Rotated Object Detection in Remote Sensing Images |
| title_full_unstemmed | Constraint Loss for Rotated Object Detection in Remote Sensing Images |
| title_short | Constraint Loss for Rotated Object Detection in Remote Sensing Images |
| title_sort | constraint loss for rotated object detection in remote sensing images |
| topic | rotated object detection remote sensing image loss functions fast convergence |
| url | https://www.mdpi.com/2072-4292/13/21/4291 |
| work_keys_str_mv | AT luyangzhang constraintlossforrotatedobjectdetectioninremotesensingimages AT haitaowang constraintlossforrotatedobjectdetectioninremotesensingimages AT lingfengwang constraintlossforrotatedobjectdetectioninremotesensingimages AT chunhongpan constraintlossforrotatedobjectdetectioninremotesensingimages AT qiangliu constraintlossforrotatedobjectdetectioninremotesensingimages AT xinyaowang constraintlossforrotatedobjectdetectioninremotesensingimages |