A Multi-Scale Convolutional Neural Network for Rotation-Invariant Recognition
The Internet of things (IoT) enables mobile devices to connect and exchange information with others over the Internet with a lot of applications in consumer, commercial, and industrial products. With the rapid development of machine learning, IoT with image recognition capability is a new research a...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2022-02-01
|
| Series: | Electronics |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2079-9292/11/4/661 |
| _version_ | 1797480828771500032 |
|---|---|
| author | Tzung-Pei Hong Ming-Jhe Hu Tang-Kai Yin Shyue-Liang Wang |
| author_facet | Tzung-Pei Hong Ming-Jhe Hu Tang-Kai Yin Shyue-Liang Wang |
| author_sort | Tzung-Pei Hong |
| collection | DOAJ |
| description | The Internet of things (IoT) enables mobile devices to connect and exchange information with others over the Internet with a lot of applications in consumer, commercial, and industrial products. With the rapid development of machine learning, IoT with image recognition capability is a new research area to assist mobile devices with processing image information. In this research, we propose the rotation-invariant multi-scale convolutional neural network (RIMS-CNN) to recognize rotated objects, which are commonly seen in real situations. Based on the dihedral group D4 transformations, the RIMS-CNN equips a CNN with multiple rotated tensors and its processing network. Furthermore, multi-scale features and shared weights are employed in the RIMS-CNN to increase performance. Compared with the data augmentation approach of using rotated images at random angles for training, our proposed method can learn inherent convolution kernels for rotational features. Experiments were conducted on the benchmark datasets: MNIST, FASHION-MNIST, CIFAR-10, and CIFAR-100. Significant improvements over the other models were achieved to show that rotational invariance could be learned. |
| first_indexed | 2024-03-09T22:06:46Z |
| format | Article |
| id | doaj.art-99759a76c27c4acdb32a1e77a8964076 |
| institution | Directory Open Access Journal |
| issn | 2079-9292 |
| language | English |
| last_indexed | 2024-03-09T22:06:46Z |
| publishDate | 2022-02-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Electronics |
| spelling | doaj.art-99759a76c27c4acdb32a1e77a89640762023-11-23T19:40:56ZengMDPI AGElectronics2079-92922022-02-0111466110.3390/electronics11040661A Multi-Scale Convolutional Neural Network for Rotation-Invariant RecognitionTzung-Pei Hong0Ming-Jhe Hu1Tang-Kai Yin2Shyue-Liang Wang3Department of Computer Science and Information Engineering, National University of Kaohsiung, Kaohsiung 811726, TaiwanDepartment of Computer Science and Information Engineering, National Cheng Kung University, Tainan 701401, TaiwanDepartment of Computer Science and Information Engineering, National University of Kaohsiung, Kaohsiung 811726, TaiwanDepartment of Information Management National, University of Kaohsiung, Kaohsiung 811726, TaiwanThe Internet of things (IoT) enables mobile devices to connect and exchange information with others over the Internet with a lot of applications in consumer, commercial, and industrial products. With the rapid development of machine learning, IoT with image recognition capability is a new research area to assist mobile devices with processing image information. In this research, we propose the rotation-invariant multi-scale convolutional neural network (RIMS-CNN) to recognize rotated objects, which are commonly seen in real situations. Based on the dihedral group D4 transformations, the RIMS-CNN equips a CNN with multiple rotated tensors and its processing network. Furthermore, multi-scale features and shared weights are employed in the RIMS-CNN to increase performance. Compared with the data augmentation approach of using rotated images at random angles for training, our proposed method can learn inherent convolution kernels for rotational features. Experiments were conducted on the benchmark datasets: MNIST, FASHION-MNIST, CIFAR-10, and CIFAR-100. Significant improvements over the other models were achieved to show that rotational invariance could be learned.https://www.mdpi.com/2079-9292/11/4/661convolutional neural networkrotational invariancemulti-scale featuredihedral groupweight sharing |
| spellingShingle | Tzung-Pei Hong Ming-Jhe Hu Tang-Kai Yin Shyue-Liang Wang A Multi-Scale Convolutional Neural Network for Rotation-Invariant Recognition Electronics convolutional neural network rotational invariance multi-scale feature dihedral group weight sharing |
| title | A Multi-Scale Convolutional Neural Network for Rotation-Invariant Recognition |
| title_full | A Multi-Scale Convolutional Neural Network for Rotation-Invariant Recognition |
| title_fullStr | A Multi-Scale Convolutional Neural Network for Rotation-Invariant Recognition |
| title_full_unstemmed | A Multi-Scale Convolutional Neural Network for Rotation-Invariant Recognition |
| title_short | A Multi-Scale Convolutional Neural Network for Rotation-Invariant Recognition |
| title_sort | multi scale convolutional neural network for rotation invariant recognition |
| topic | convolutional neural network rotational invariance multi-scale feature dihedral group weight sharing |
| url | https://www.mdpi.com/2079-9292/11/4/661 |
| work_keys_str_mv | AT tzungpeihong amultiscaleconvolutionalneuralnetworkforrotationinvariantrecognition AT mingjhehu amultiscaleconvolutionalneuralnetworkforrotationinvariantrecognition AT tangkaiyin amultiscaleconvolutionalneuralnetworkforrotationinvariantrecognition AT shyueliangwang amultiscaleconvolutionalneuralnetworkforrotationinvariantrecognition AT tzungpeihong multiscaleconvolutionalneuralnetworkforrotationinvariantrecognition AT mingjhehu multiscaleconvolutionalneuralnetworkforrotationinvariantrecognition AT tangkaiyin multiscaleconvolutionalneuralnetworkforrotationinvariantrecognition AT shyueliangwang multiscaleconvolutionalneuralnetworkforrotationinvariantrecognition |