A Fine-Grained Bird Classification Method Based on Attention and Decoupled Knowledge Distillation

Wang, Kang; Yang, Feng; Chen, Zhibo; Chen, Yixin; Zhang, Ying

doi:10.3390/ani13020264

Cited by 8 publications

(6 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…where α and β are hyperparameters that need to be manually set. In CIFAR-100 and ImageNet classification tasks, using the DKD strategy yields better results compared to using KD strategy based on intermediate layer features [50]- [52].…”

Section: Related Workmentioning

confidence: 99%

Decoupled Knowledge Distillation via Spatial Feature Blurring for Hyperspectral Image Classification

Xie,

Zhang,

Jiao

et al. 2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Section: Related Workmentioning

confidence: 99%

Decoupled Knowledge Distillation via Spatial Feature Blurring for Hyperspectral Image Classification

Xie,

Zhang,

Jiao

et al. 2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

“…The FasterNet module also reduces the overall computational speed [25]. In contrast, they are placed after each intermediate PWConv to maintain feature diversity and achieve lower latency.…”

Section: Faster-block Module In Fasternetmentioning

confidence: 99%

RTDETR-Refa: a real-time detection method for multi-breed classiffcation of cattle

Li,

Fang,

Zhao

2024

Preprint

View full text Add to dashboard Cite

In the farming industry in order to cope with the problems such as the complex environment of the pasture and dense targets, etc., which lead to increased difficulty in recognition, so as to quickly classify and automatically identify cattle breeds and improve the accuracy. In this paper, an RTDETR-Refa (Rep- Conv Efficient Faster Attention) model is proposed for classification and identification of cattle breeds. Firstly, some novel improvements have been made to ResNet18. Replacing the 1×1 convolution in Faster-Block with the reparameterised RepConv, and adding the Efficient Multiscale Attention Module (EMA) in front of the global average pooling layer in Faster-Block to enhance the transformation and classification of features, Finally, the 4-layer BasicBlock after the 3 convolutional layers in the resnet18-backbone is replaced by the improved Faster-Block.Finally, the results of the training tests of the RTDETR-Refa model are compared with other classical models: the YOLO family of models: the YOLOv5m, YOLOv6, YOLOv8m, YOLOv9 and the state-of-the-art (SOTA) CNN backbone networks EfficientViT, FasterNet, UniRepLKNet, TransNeXt verifying their superiority. The average accuracy of the RTDETR-Refa model on the cattle classification training set is 91.6%, which is 0.8% higher than that of ResNet18 and 0.9-5.2% higher than that of other classical models. The experimental results show that the RTDETR-Refa model proposed in this paper is capable of identifying and classifying cattle of different breeds while ensuring similar detection speeds, demonstrating the feasibility of convolutional neural networks in breed identification and classification.

show abstract

“…To validate the efficacy of our proposed model, we compared it with established fine-grained classification approaches, encompassing both strongly supervised (HSnet [38], Mask-CNN [39]) and weakly supervised (RA-CNN [40], ADKD [41]) techniques.…”

Section: State-of-the-art Comparisonmentioning

confidence: 99%

Automatic Bird Species Recognition from Images with Feature Enhancement and Contrastive Learning

Yang,

Shen,

2024

Applied Sciences

Self Cite

View full text Add to dashboard Cite

Accurate bird species recognition is crucial for ecological conservation, wildlife monitoring, and biological research, yet it poses significant challenges due to the high variability within species and the subtle similarities between different species. This paper introduces an automatic bird species recognition method from images that leverages feature enhancement and contrast learning to address these challenges. Our method incorporates a multi-scale feature fusion module to comprehensively capture information from bird images across diverse scales and perspectives. Additionally, an attention feature enhancement module is integrated to address noise and occlusion within images, thus enhancing the model’s robustness. Furthermore, employing a siamese network architecture allows effective learning of common features within instances of the same class and distinctions between different bird species. Evaluated on the CUB200-2011 dataset, our proposed method achieves state-of-the-art performance, surpassing existing methods with an accuracy of 91.3% and F1 score of 90.6%. Moreover, our approach showcases a notable advantage in scenarios with limited training data. When utilizing only 5% of the training data, our model still achieves a recognition accuracy of 65.2%, which is significantly higher than existing methods under similar data constraints. Notably, our model exhibits faster execution times compared to existing methods, rendering it suitable for real-time applications.

show abstract

A Fine-Grained Bird Classification Method Based on Attention and Decoupled Knowledge Distillation

Cited by 8 publications

References 33 publications

Decoupled Knowledge Distillation via Spatial Feature Blurring for Hyperspectral Image Classification

Decoupled Knowledge Distillation via Spatial Feature Blurring for Hyperspectral Image Classification

RTDETR-Refa: a real-time detection method for multi-breed classiffcation of cattle

Automatic Bird Species Recognition from Images with Feature Enhancement and Contrastive Learning

Contact Info

Product

Resources

About