基于对抗学习的野生动物图像域适应识别方法

doi:10.11707/j.1001-7488.LYKX20240437

Abstract

Abstract:

Objective: To address the domain shift issues caused by different spatiotemporal scenarios and species variations, this study proposes a domain adaptation recognition method for wildlife images based on adversarial learning. The aim of this study is to enhance the generalization performance of unlabeled wildlife species recognition in complex field environment, thereby providing a robust theoretical foundation for the classification of wildlife in open environment. Method: An adversarial learning network model was constructed by utilizing wildlife image category information as conditional inputs. Batch spectral penalty constraints and mixup feature alignment methods were employed to mitigate distribution discrepancies among wildlife images across different spatiotemporal scenarios and species variations. This framework established a feature alignment model based on adversarial learning to enhance image recognition performance. Result: The proposed method was trained and evaluated on domain adaptation datasets containing 8 and 11 wildlife species, respectively. Compared to baseline methods combining adversarial learning and feature alignment, the average recognition accuracy of the proposed method increased by 3.3% and 14.0%, the precision was improved by 3.3% and 20.6%, the recall rate was improved by 3.5% and 20.5%, and the F1-score was improved by 3.6% and 5.1%, respectively. The wild animal image domain adaptation model based on adversarial learning significantly improved performance for wildlife recognition in cross-domain scenarios. Conclusion: The wildlife category information used as conditional inputs in the adversarial learning network can provide multimodal structural information of wildlife images, which enables the proposed method in this study to better understand the inter-image relationships and improve domain adaptation learning capabilities. The better the alignment of image features between the training and testing sets, the better the image recognition performance of the testing set. This study provides novel insights and methods for cross-domain wildlife image recognition research.

Key words: wildlife, recognition method, adversarial learning, feature alignment, domain adaptation

CLC Number:

S718.6
S862

Enting Zhao,Changchun Zhang,Haitao Zhao,Junguo Zhang. A Recognition Method of Domain Adaptation for Wildlife Images Based on Adversarial Learning[J]. Scientia Silvae Sinicae, 2025, 61(4): 1-8.

Figures/Tables 9

Fig.1

Table 1

Table 2

Table 3

Table 4

Fig.2

Table 5

Table 6

Fig.3

References 0

	程浙安. 2019. 基于深度卷积神经网络的内蒙古地区陆生野生动物自动识别. 北京: 北京林业大学.
	Cheng Z A. 2019. Automatic recognition of terrestrial wildlife in inner mongolia based on deep convolution neural network. Beijing: Beijing Foresty University. ［in Chinese］
	宫一男, 谭孟雨, 王　震, 等. 基于深度学习的红外相机动物影像人工智能识别: 以东北虎豹国家公园为例. 兽类学报, 2019, 39 (4): 458- 465.
	Gong Y N, Tan M Y, Wang Z, et al. Al recognition of infrared camera image of wild animals based on deep learning: Northeast Tiger and Leopard National Park for example. Acta Theriologoca Sinica, 2019, 39 (4): 458- 465.
	李安琪. 2020. 基于卷积神经网络的野生动物监测图像自动识别方法研究. 北京: 北京林业大学.
	Li A Q. 2020. Research on automatic recognition method of wildlife monitoring images based on convolutional neural network. Beijing: Beijing Forestry University. ［in Chinese］
	齐建东, 马鐘添, 张德怀, 等. 基于BS-ResNeXt-50的密云地区野生动物图像识别. 林业科学, 2023, 59 (8): 112- 122. doi: 10.11707/j.1001-7488.LYKX20220378
	Qi J D, Ma Z T, Zhang D H, et al. Wildlife image recognition in Miyun District based on BS-ResNeXt-50. Scientia Silvae Sinicae, 2023, 59 (8): 112- 122. doi: 10.11707/j.1001-7488.LYKX20220378
	肖治术, 肖文宏, 王天明, 等. 中国野生动物红外相机监测与研究: 现状及未来. 生物多样性, 2022, 30 (10): 234- 259. doi: 10.17520/biods.2022451
	Xiao Z S, Xiao W H, Wang T M, et al. Wildlife monitoring and research using camera-trapping technology across China: the current status and future issues. Biodiversity Science, 2022, 30 (10): 234- 259. doi: 10.17520/biods.2022451
	张长春, 李大方, 张军国. 基于Wasserstein距离和相关对齐迁移学习的野生动物图像识别方法. 林业科学, 2024, 60 (8): 25- 32. doi: 10.11707/j.1001-7488.LYKX20230399
	Zhang C C, Li D F, Zhang J G. Wildlife images recognition method based on Wasserstein distance and correlation alignment transfer learning. Scientia Silvae Sinicae, 2024, 60 (8): 25- 32. doi: 10.11707/j.1001-7488.LYKX20230399
	张军国, 程浙安, 胡春鹤, 等. 野生动物监测光照自适应Retinex图像增强算法. 农业工程学报, 2018, 34 (15): 183- 189. doi: 10.11975/j.issn.1002-6819.2018.15.023
	Zhang J G, Cheng Z A, Hu C H, et al. Adaptive image enhancement algorithm for wild animal monitoring based on Retinex theory. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2018, 34 (15): 183- 189. doi: 10.11975/j.issn.1002-6819.2018.15.023
	Ahn S, Kim S, Jeong D. Unsupervised domain adaptation for mitigating sensor variability and interspecies heterogeneity in animal activity recognition. Animals, 2023, 13 (20): 3276. doi: 10.3390/ani13203276
	Bakana S, Zhang Y, Twala B. WildARe-YOLO: a lightweight and efficient wild animal recognition model. Ecological Informatics, 2024, 80, 102541. doi: 10.1016/j.ecoinf.2024.102541
	Bohdal O, Li D, Hu S, et al. 2024. Feed-forward latent domain adaptation. IEEE Winter Conference on Applications of Computer Vision, 8490−8499.
	Chen X, Wang S, Long M, et al. 2019. Transferability vs. discriminability: Batch spectral penalization for adversarial domain adaptation. International Conference on Machine Learning, 1081−1090.
	He K, Zhang X, Ren S, et al. 2016. Deep residual learning for image recognition. IEEE Conference on Computer Vision and Pattern Recognition, 770−778.
	Henrich M, Burgueño M, Hoyer J, et al. A semi-automated camera trap distance sampling approach for population density estimation. Remote Sensing in Ecology and Conservation, 2024, 10 (2): 156- 171. doi: 10.1002/rse2.362
	Ganin Y, Ustinova E, Ajakan H, et al. Domain-adversarial training of neural networks. Journal of Machine Learning Research, 2016, 17 (1): 2096- 2030.
	Long M, Cao Z, Wang J, et al. Conditional adversarial domain adaptation. Advances in Neural Information Processing Systems, 2018, 31, 1640- 1650.
	Ma Z, Dong Y, Xia Y, et al. Wildlife real-time detection in complex forest scenes based on YOLOv5s deep learning network. Remote Sensing, 2024, 16 (8): 1350. doi: 10.3390/rs16081350
	Na J, Jung H, Chang H J, et al. 2021. Fixbi: Bridging domain spaces for unsupervised domain adaptation. IEEE Conference on Computer Vision and Pattern Recognition, 1094−1103.
	Peng X, Bai Q, Xia X, et al. 2019. Moment matching for multi-source domain adaptation. IEEE International Conference on Computer Vision, 1406−1415.
	Simonyan K, Zisserman A. 2015. Very deep convolutional networks for large-scale image recognition. International Conference on Learning Representations, 1−14.
	Sun B, Saenko K. 2016. Deep coral: correlation alignment for deep domain adaptation. European Conference on Computer Vision, 443−450.
	Tabak M, Norouzzadeh M, Wolfson D, et al. Machine learning to classify animal species in camera trap images: applications in ecology. Methods in Ecology and Evolution, 2019, 10, 585- 590. doi: 10.1111/2041-210X.13120
	Whytock R, Świeżewski J, Zwerts J, et al. Robust ecological analysis of camera trap data labelled by a machine learning model. Methods in Ecology and Evolution, 2021, 12 (6): 1080- 1092. doi: 10.1111/2041-210X.13576
	Yousif H, Kays R, He Z. 2019. Dynamic programming selection of object proposals for sequence-level animal species classification in the wild. IEEE Transactions on Circuits and Systems for Video Technology, 20.
	Zhang C C, Zhang J G. DJAN: Deep joint adaptation network for wildlife image recognition. Animals, 2023, 13 (21): 3333. doi: 10.3390/ani13213333
	Zhou L H, Ye M, Li X P, et al. Disentanglement then reconstruction: Unsupervised domain adaptation by twice distribution alignments. Expert Systems with Applications, 2024, 237, 121498. doi: 10.1016/j.eswa.2023.121498

方法 Methods	识别任务Recognition task		平均准确率 Average accuracy
方法 Methods	R→C 准确率Accuracy	C→R准确率Accuracy	平均准确率 Average accuracy
ResNet50	81.9	68.8	75.4
DCORAL	89.5	84.0	86.8
CDAN	88.3	91.0	89.7
DJAN	92.7	93.8	93.3
本研究方法Ours	97.8	95.3	96.6

方法Methods	R→C			C→R
方法Methods	精确率 Precision	召回率 Recall	F1	精确率Precision	召回率Recall	F1
ResNet50	81.3	76.8	77.4	69.6	70.4	68.1
DCORAL	89.1	89.5	88.9	84.6	84.5	84.3
CDAN	90.4	88.3	88.4	91.0	91.0	90.9
DJAN	93.3	92.7	92.7	92.9	92.8	92.7
本研究方法Ours	97.7	97.8	97.7	95.0	94.7	94.8

物种 Species	R→C准确率 Accuracy
物种 Species	ResNet50	DCORAL	CDAN	DJAN	本研究方法Ours
熊Bear	93.2	79.7	93.5	93.2	98.8
象 Elephant	96.3	97.5	99.0	99.2	99.2
长颈鹿Giraffe	98.3	95.8	96.0	98.1	98.3
狮子Lion	60.0	81.3	79.7	86.9	94.9
浣熊Raccoon	47.0	85.5	87.2	91.4	95.8
松鼠Squirrel	65.5	81.8	96.2	96.0	96.2
虎Tiger	54.9	94.8	55.3	77.5	99.8
斑马Zebra	99.4	99.4	99.4	99.6	99.4

方法 Methods	识别任务Recognition task		平均准确率 Average accuracy	识别任务Recognition task		平均F1值 Average F1
方法 Methods	ES→NS 准确率Accuracy	NS→ES 准确率Accuracy	平均准确率 Average accuracy	ES→NS F1	NS→ES F1	平均F1值 Average F1
ResNet50	41.4	55.4	48.4	25.6	47.9	36.8
DCROAL	44.1	55.1	49.6	33.7	43.8	38.8
CDAN	45.5	56.2	50.9	35.4	43.2	39.3
DJAN	48.8	67.5	58.2	35.4	61.6	48.5
本研究方法Ours	65.9	78.5	72.2	46.9	60.2	53.6

方法 Methods	识别任务Recognition task		平均准确率 Average accuracy
方法 Methods	ES→NS 准确率Accuracy	NS→ES 准确率Accuracy	平均准确率 Average accuracy
VGG16	42.0	55.0	48.5
DCORAL	40.8	53.3	47.1
CDAN	43.8	56.5	50.2
WDCA-TLNet	44.9	58.3	51.6
本研究方法Ours	70.0	85.2	77.6

A Recognition Method of Domain Adaptation for Wildlife Images Based on Adversarial Learning

RichHTML

PDF (PC)

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 9

References 0

Related Articles 8

Recommended Articles

Metrics

Comments

方法 Methods	识别任务 Recognition task		平均准确率 Average accuracy
方法 Methods	R→C准确率Accuracy	C→R准确率Accuracy	平均准确率 Average accuracy
ResNet50	81.9	68.8	75.4
－mixup	90.9	75.3	83.1
－bsp	96.1	92.1	94.1
－CDAN	97.3	93.3	95.3
本研究方法Ours	97.8	95.3	96.6

[1]	Jiandong Qi,Shangzi Zheng,Ziyi Chen,Zhongtian Ma. Wildlife Image Recognition of Infrared Cameras in Beijing Area Based on an Improvement ConvNeXt Model [J]. Scientia Silvae Sinicae, 2024, 60(8): 33-45.
[2]	Wenhan Yang,Tianyu Liu,Junchi Zhou,Wenwu Hu,Ping Jiang. CNN-Swin Transformer Detection Algorithm of Forest Wildlife Images Based on Improved YOLOv5s [J]. Scientia Silvae Sinicae, 2024, 60(3): 121-130.
[3]	Jiandong Qi,Zhongtian Ma,Dehuai Zhang,Yun Tian. Wildlife Image Recognition in Miyun District Based on BS-ResNeXt-50 [J]. Scientia Silvae Sinicae, 2023, 59(8): 112-122.
[4]	Junfeng Chen,Yi Xie. Wildlife Accident, Compensation for Damage Caused by Wildlife and Farmers’ Willingness to Protect Wildlife [J]. Scientia Silvae Sinicae, 2023, 59(12): 152-166.
[5]	Yuxuan Hu,Junfeng Chen,Yi Xie. Measures for Governing Human-Elephant Conflicts Based on Choice Experiment of Farmers in Xishuangbanna [J]. Scientia Silvae Sinicae, 2023, 59(10): 162-170.
[6]	Jinhua Mo,Jia Li,Fang Liu,Xiaoguan Li,Diqiang Li. A Survey of Mammals and Birds Diversity in Jianfengling District of Hainan Province by Using Camera-Trapping [J]. Scientia Silvae Sinicae, 2019, 55(10): 203-210.
[7]	Chen Lixia, Liu Huajin, Liu Yulin, Yang Peiyu, Zhang Guogang, Lu Jun. Analysis on the Variation of Waterbird Communities in Different Habitats of Khanka Lake in China [J]. Scientia Silvae Sinicae, 2019, 55(1): 56-65.
[8]	Li Yongcheng;Sun Heting;Liu Feng. The Role of Wildlife in the Occurrence and Prevalence of Zoonosis [J]. Scientia Silvae Sinicae, 2006, 42(5): 120-126.