改进ConvNeXt网络的树种识别方法

doi:10.11707/j.1001-7488.LYKX20240121

摘要/Abstract

摘要：

目的: 为提高树种识别工作的效率和准确率，提出一种利用迁移学习策略并引入SimAM注意力机制和ECA通道注意力机制的ConvNeXt树种识别模型。方法: 以12种常见树种的树皮图像为研究对象，通过传统数据增强方法对数据进行扩充，防止模型过拟合。使用SimAM和ECA通道注意力机制构建以ConvNeXt为基础的改进网络，增强特征提取的SA-ConvNeXt、增强重要特征权重的E-ConvNeXt、结合两者的ES-ConvNeXt，测试数据集在增强前后对ES-ConvNeXt网络准确率的影响。使用ResNet34、ResNet50、GoogLeNet、Swin Transformer、DenseNet121和ConvNeXt网络，与ES-ConvNeXt模型识别效果进行比较。结果: SA-ConvNeXt和E-ConvNeXt准确率分别达到(95.14±0.42)%、(96.085±0.235)%，ES-ConvNeXt在增强后数据集测试的准确率达到(97.445±0.635)%，对单一树种识别准确率均超过93%，最高类别准确率达到99.79%，为最优方案。经数据增强后进行训练的模型与使用原始数据进行训练的模型相比，其验证集的准确率和损失值，无论是收敛速度还是最终稳定值都是最优。数据集相同时，使用ResNet34、ResNet50、GoogLeNet、Swin Transformer、DenseNet121和ConvNeXt网络的识别准确率，分别为92.74%、94.47%、90.52%、92.85%、70.38%、94.72%，均低于新改进模型ES-ConvNeXt（97.81%），进一步说明了改进后的ES-ConvNeXt模型的有效性。结论: 数据增强对模型准确率提升有效，在数据增强后的数据集上，改进后的ES-ConvNeXt模型与其他模型相比可以更加准确地完成树种分类任务，在不同树种上也有较好的泛化能力。

关键词: 树种识别, ConvNeXt, SimAM注意力机制, ECA通道注意力机制

Abstract:

Objective: In this study, an improved tree species recognition model of ConvNeXt network was proposed by using a transfer learning strategy and introducing the SimAM attention module and ECA channel attention mechanism, so as to improve the efficiency and accuracy of tree species recognition work and solve the difficulties encountered in the recognition work. Method: The bark images of common 12 tree species were used as the research object, and the data were expanded by traditional data enhancement methods to prevent model overfitting. An improved ConvNeXt-based network was constructed using SimAM and ECA channel attention mechanisms: SA-ConvNeXt for enhanced feature extraction, E-ConvNeXt for enhanced weighting of important features, and ES-ConvNeXt combining the two. The effect of the dataset on the accuracy of the ES-ConvNeXt network before and after enhancement was tested. The recognition effects with the ES-ConvNeXt model were compared by using the Resnet34, Rennet50, GoogLeNet, Swin Transformer, Densenet121, and ConvNeXt networks. Result: SA-ConvNeXt and E-ConvNeXt achieved 95.14%±0.42% and 96.085%±0.235% accuracy, respectively. ES-ConvNeXt, which incorporates SimAm and ECA attention modules, achieved an accuracy of 97.445%±0.635% for the test on the augmented dataset, its recognition accuracy for a single tree species exceeded 93%, and the highest category accuracy reached 99.79%, making it the optimal solution. The model trained with expanded data had optimal accuracy and loss values for the validation set both in terms of speed of convergence and final stabilized values compared to the model trained using the original data. With the same dataset, the recognition accuracies using Resnet34, Rennet50, GoogLeNet, Swin Transformer, Densenet121, and ConvNeXt networks were 92.74%, 94.47%, 90.52%, 92.85%, 70.38%, and 94.72%, respectively, which were all lower than the 97.81% obtained by the new improved model (ES-ConvNeXt model), further illustrating the effectiveness of the improved ES-ConvNeXt model. Conclusion: Data enhancement is effective for model accuracy improvement, and on the data-enhanced dataset, the improved ES-ConvNeXt model can perform the tree classification task more accurately compared to the other models, and it also has better generalization ability on different tree species.

Key words: tree species recognition, ConvNeXt, SimAM attention mechanism, ECA channel attention mechanism

中图分类号:

TP391.4

杨兵兵,许杰. 改进ConvNeXt网络的树种识别方法[J]. 林业科学, 2025, 61(2): 31-39.

Bingbing Yang,Jie Xu. Tree Species Recognition Based on Improved ConvNeXt Network[J]. Scientia Silvae Sinicae, 2025, 61(2): 31-39.

图/表 11

图1

图2

表1

图3

图4

表2

图5

图6

表3

表4

图7

参考文献 0

	刘　骥, 曹凤莲, 甘林昊. 基于叶片形状特征的植物识别方法. 计算机应用, 2016, 36 (S2): 200- 202,226.
	Liu J, Cao F L, Gan L H. Plant identification method based on leaf shape features. Journal of Computer Applications, 2016, 36 (S2): 200- 202,226.
	苏　彤, 许　杰. 基于生成对抗网络的树种识别方法. 林业科学, 2024, 60 (2): 97- 105.
	Su T, Xu J. Tree species identification method based on generative adversarial network. Scientia Silvae Sinicae, 2024, 60 (2): 97- 105.
	王丽君, 淮永建, 彭月橙. 基于叶片图像多特征融合的观叶植物种类识别. 北京林业大学学报, 2015, 37 (1): 55- 61.
	Wang L J, Huai Y J, Peng Y C. Method of identification of foliage from plants based on extraction of multiple features of leaf images. Journal of Beijing Forestry University, 2015, 37 (1): 55- 61.
	张善文, 张晴晴, 齐国红. 基于Fourier描述子和LBP相结合的植物叶片识别方法. 江苏农业科学, 2019, 47 (14): 273- 276.
	Zhang S W, Zhang Q Q, Qi G H. A plant leaf recognition method based on Fourier descriptor and local binary pattern. Jiangsu Agricultural Sciences, 2019, 47 (14): 273- 276.
	赵　霖, 张晓丽, 吴艳双, 等. 面向机载高光谱数据的3D-CNN亚热带森林树种分类. 林业科学, 2020, 56 (11): 97- 107.
	Zhao L, Zhang X L, Wu Y S, et al. Subtropical forest tree species classification based on 3D-CNN for airborne hyperspectral data. Scientia Silvae Sinicae, 2020, 56 (11): 97- 107.
	朱　莉, 宋绪秋, 邢　鑫, 等. 基于改进ResNet34网络的树种识别研究. 国外电子测量技术, 2022, 41 (7): 119- 125.
	Zhu L, Song X Q, Xing X, et al. Research on tree species identification based on improved ResNet34 network. Foreign Electronic Measurement Technology, 2022, 41 (7): 119- 125.
	Faizal S. Automated identification of tree species by bark texture classification using convolutional neural networks. International Journal for Research in Applied Science and Engineering Technology, 2022, 10 (9): 1384- 1392. doi: 10.22214/ijraset.2022.46846
	He K M, Zhang X Y, Ren S Q, et al. 2016. Deep residual learning for image recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 770–778.
	Huang G, Liu Z, van Der Maaten L, et al. Densely connected convolutional networks. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017, Honolulu, HI,4700- 4708.
	Liu Z, Lin Y T, Cao Y, et al. 2021. Swin transformer: hierarchical vision transformer using shifted windows. 2021 IEEE/CVF International Conference on Computer Vision (ICCV), Montreal, QC, Canada, 10012–10022.
	Liu Z, Mao H Z, Wu C Y, et al. 2022. A ConvNet for the 2020s. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, LA, USA, 11976–11786.
	Lü Z M, Zhang Z B. Research on plant leaf recognition method based on multi-feature fusion in different partition blocks. Digital Signal Processing, 2023, 134, 103907. doi: 10.1016/j.dsp.2023.103907
	Miyoshi G T, dos Santos Arruda M, Osco L P, et al. A novel deep learning method to identify single tree species in UAV-based hyperspectral images. Remote Sensing, 2020, 12 (8): 1294. doi: 10.3390/rs12081294
	Selvaraju R R, Cogswell M, Das A, et al. Grad-CAM: visual explanations from deep networks via gradient-based localization. International Journal of Computer Vision, 2020, 128 (2): 336- 359. doi: 10.1007/s11263-019-01228-7
	Sun Y, Liu Y, Wang G, et al. 2017. Deep learning for plant identification in natural environment. Computational Intelligence and Neuroscience, 7361042.
	Szegedy C, Liu W, Jia Y Q, et al. 2015. Going deeper with convolutions. 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Boston: MA,USA,1−9.
	Wang Q L, Wu B G, Zhu P F, et al. 2020. ECA-Net: efficient channel attention for deep convolutional neural networks. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle: WA,USA,11534−11542.
	Wu F Y, Gazo R, Benes B, et al. Deep BarkID: a portable tree bark identification system by knowledge distillation. European Journal of Forest Research, 2021, 140 (6): 1391- 1399. doi: 10.1007/s10342-021-01407-7
	Yang L, Zhang R Y, Li L, et al. 2021. Simam: a simple, parameter-free attention module for convolutional neural networks. Proceedings of the International Conference on Machine Learning, 11863–11874.
	Zamir A R, Sax A, Shen W, et al. 2018. Taskonomy: disentangling task transfer learning. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, 3712–3722.
	Zhong H, Lin W S, Liu H R, et al. Identification of tree species based on the fusion of UAV hyperspectral image and LiDAR data in a coniferous and broad-leaved mixed forest in northeast China. Frontiers in Plant Science, 2022, 13, 964769. doi: 10.3389/fpls.2022.964769

结构 Structure	输入 Input	卷积核及步距 Convolution kernel and step size	输出 Output
卷积层：ES-Block层1 Convolution layer： ES-Block layer 1	224×224×3	4×4，s4	56×56×96
	56×56×96	d7×7，s1	56×56×96
	56×56×96	1×1，s1	56×56×384
	56×56×384	1×1，s1	56×56×96
下采样：ES-Block层2 Downsampling： ES-Block layer 1	56×56×96	2×2，s2	28×28×192
	56×56×192	d7×7，s1	28×28×192
	28×28×192	1×1，s1	28×28×768
	28×28×768	1×1，s1	28×28×192
下采样：ES-Block层3 Downsampling： ES-Block layer 3	28×28×192	2×2，s2	14×14×384
	14×14×384	d7×7，s1	14×14×384
	14×14×384	1×1，s1	14×14×1536
	14×14×1536	1×1，s1	14×14×384
下采样：ES-Block层4 Downsampling： ES-Block layer 4	14×14×384	2×2，s2	7×7×768
	7×7×768	d7×7，s1	7×7×768
	7×7×768	1×1，s1	7×7×3072
	7×7×3072	1×1，s1	7×7×768

物种 Species	模型准确率Model accuracy（%）
物种 Species	ConvNeXt	SA-ConvNeXt	E-ConvNeXt	ES-ConvNeXt
梣叶槭Acer negundo	82.75	93.15	94.46	95.72
枫香树Liquidambar formosana	92.48	95.27	95.88	97.31
洋槐树Robinia pseudoacacia	92.44	97.25	96.21	97.47
银杏Ginkgo biloba	80.77	94.58	95.64	97.72
复羽叶栾树Koelreuteria bipinnata	87.61	94.80	93.51	96.84
家榆Ulmus pumila	89.72	95.90	96.29	95.96
女贞Ligustrum lucidum	88.38	93.44	93.92	93.36
山杨Populus davidiana	99.36	99.79	100.00	99.79
水曲柳Fraxinus mandshurica	90.70	91.94	94.56	96.63
乌桕Sapium sebiferum	82.85	93.53	95.44	97.47
梧桐Firmiana simplex	99.58	99.38	99.17	99.58
喜树Camptotheca acuminata	88.16	94.85	95.17	97.27

数据集 Database	准确率 Accuracy	精确率 Precision	召回率 Recall	特异性 Specificity
原始数据 Original data	96.76	96.23	96.07	99.71
增强后数据 After data enhancement	97.81	97.82	97.81	99.80

模型 Model	准确率 Accuracy	精确率 Precision	召回率 Recall	特异性 Specificity
Swin-T	92.85	92.91	92.86	99.36
ResNet34	92.74	92.84	92.73	99.34
ResNet50	94.47	94.53	94.48	99.50
GoogLeNet	90.52	90.53	90.52	99.14
DenseNet121	70.38	70.98	70.40	97.54
ConvNeXt	94.72	94.74	94.72	99.53
ES-ConvNeXt	97.81	97.82	97.81	99.80

[1]	齐建东,郑尚姿,陈子仪,马鐘添. 基于ConvNeXt的北京地区红外相机野生动物图像识别改进模型构建[J]. 林业科学, 2024, 60(8): 33-45.
[2]	苏彤, 许杰. 基于生成对抗网络的树种识别方法[J]. 林业科学, 2024, 60(2): 97-105.
[3]	潘玺,李康,杨忠. 基于卷积神经网络的近红外光谱与数字图像特征信息融合木材树种识别[J]. 林业科学, 2024, 60(12): 136-145.
[4]	岳巍,李世明,李增元,刘清旺,庞勇,斯林. 基于多时相Sentinel-2影像和SNIC分割算法的优势树种识别[J]. 林业科学, 2022, 58(9): 60-69.
[5]	黄翀,张晨晨,刘庆生,李贺,杨晓梅,刘高焕. 结合光学与雷达影像多特征的热带典型人工林树种精细识别[J]. 林业科学, 2021, 57(7): 80-91.
[6]	栗旭升,李虎,陈冬花,刘玉锋,刘赛赛,刘聪芳,胡国庆. 联合GF-5与GF-6卫星数据的多分类器组合亚热带树种识别[J]. 林业科学, 2020, 56(10): 93-104.