基于 Gabor特征提取和 MobileNetV2在云南天牛识别中的应用

Application of Gabor Feature Extraction and MobileNetV2 in Yunnan Longicorn Beetle Identification

摘要

天牛是对林木健康构成重大威胁的害虫，其分类和识别对于生态学、农业和环境保护至关重要。天牛分类方法中，通常采用传统的形态学分类法，但存在效率较低等问题。本文以云南常见的10种天牛为研究对象，采集天牛标本和自然背景图片作为数据集，通过2D Gabor滤波器对天牛图像进行特征提取，引入轻量化的迁移学习模型MobileNetV2进行分类。实验对比了局部二值模式（LBP）、灰度共生矩阵（GLCM）、尺度不变特征变换（SIFT）提取特征方法与支持向量机（SVM）、随机森林（RF）分类器，以及VGG16、ResNet101、InceptionV3和MobileNetV2模型的组合性能。结果表明，LBP_RF和GLCM_RF的准确率分别为61.93%和67.93%；原始数据集（SWFU LHB 10）在VGG16、ResNet101、InceptionV3和MobileNetV2上的准确率分别为70.90%、41.53%、76.10%和83.07%；SIFT特征提取后模型性能普遍下降；而在Gabor特征提取的基础上，MobileNetV2模型识别准确率提升至98.94%，F1-score达98.80%，因此，采用2D Gabor滤波器结合MobileNetV2模型进行天牛分类的方法，在特征提取和模型训练方面均显著优于其他方法，为天牛识别提供了有效的解决方案，对天牛的综合防治具有重要意义，也可为相关行业领域提供参考。

Abstract

Longicorn beetles pose a significant threat to forest health, making their classification and identification crucial for ecology, agriculture, and environmental protection. Typically, existing classification methods rely on traditional morphological taxonomy, which suffers from low efficiency and limited accuracy. This paper focuses on 10 common species of longhorn beetles in Yunnan Province, utilizing a dataset comprising collected specimens and images with natural backgrounds. It applies a 2D Gabor filter to extract image texture features, and introduces the lightweight transfer learning model MobileNetV2 for classification. The study compares the performance of feature extraction methods, including Local Binary Pattern (LBP), Gray Level Co-occurrence Matrix (GLCM), and Scale Invariant Feature Transform (SIFT), combined with classifiers such as Support Vector Machine (SVM), Random Forest (RF), as well as models like VGG16, ResNet101, InceptionV3, and MobileNetV2. The Results indicate that the classification accuracies of LBP_RF and GLCM_RF are 61.93% and 67.93%, respectively. The accuracy of the original dataset (SWFU LHB 10) on VGG16, ResNet101, InceptionV3, and MobileNetV2 reaches 70.90%, 41.53%, 76.10%, and 83.07%, respectively. However, performance declines after applying SIFT features. In contrast, combining Gabor features with MobileNetV2 significantly improves the classification accuracy to 98.94%, with an F1-score of 98.80%. Therefore, the proposed method based on 2D Gabor filtering and MobileNetV2 significantly outperforms other approaches in both feature extraction and model training. It provides an effective solution for longhorn beetle identification.

关键词

天牛识别 / 2D Gabor / 特征提取 / MobileNetV2 / 深度学习

Key words

Beetle classification / 2D Gabor / feature extraction / MobileNetV2 / deep learning

参考文献

[1] Linsley E G . Ecology of cerambycidae[J]. Annual Review of Entomology, 1959, 4(01): 99-138.

[2] 李金兰. 重庆市天牛总科昆虫分类研究[D].重庆:重庆师范大学, 2023.

[3] 李丽莎. 云南天牛[M].昆明:云南科技出版社, 2009: 3.

[4] 李丹丹. 云南林地森林面积均居全国第二位[N/OL]. 昆明日报, 2022-12-01(004)[2025-1-12]. https://kmrb.kmzscc.com/.

[5] 刘梦盈. 云南松对云南松小蠹危害的光谱响应及其受害监测研究[D].北京:中国林业科学研究院, 2022.

[6] 云南省林业和草原局. 昆明林草原局推进林业有害生物防治工作[EB/OL].(2020-07-29)[2025-1-12]. https://lcj.yn.gov.cn/html/2020/kunming_0729/57325.htm.

[7] 谭亮魁, 王文凯. 生物技术在天牛分类中的应用[J]. 中国农学通报, 2007(12): 53-57.

[8] 付甫永, 司徒春南, 路纪芳, 等. 贵州省核桃蛀干害虫白条天牛的分类特征分析研究[J]. 绿色科技, 2016(17): 81-83.

[9] 满庆丽. 天牛图像的特征提取和识别算法的研究[D].哈尔滨:东北林业大学, 2013.

[10] 夏令. 天牛幼虫蛀木振动信号的降噪及特征识别[D].北京:北京交通大学, 2022.

[11] 慕君林, 马博, 王云飞, 等. 基于深度学习的农作物病虫害检测算法综述[J]. 农业机械学报, 2023, 54(S2): 301-313.

[12] 彭鑫. 基于深度学习的农业害虫识别方法研究[D].吉林:吉林化工学院, 2024.

[13] 孙丰刚, 王建丽, 季英超, 等. 基于机器视觉的松林天牛远程智能监测系统[J]. 农业工程学报, 2023, 39(17): 190-198.

[14] Marcelja S . Mathematical description of the responses of simple cortical cells[J]. Journal of the Optical Society of America, 1980, 70(11): 1297-1300.

[15] Daugman J G . Two-dimensional spectral analysis of cortical receptive field profiles[J]. Vision Research, 1980, 20(10): 847-856.

[16] 杨旺功, 淮永建, 张福泉. 基于Gabor及深度神经网络的葡萄种子分类[J]. 电子科技大学学报, 2020, 49(01): 131-138.

[17] Hu X , Wang X , Meng F , et al. Gabor-CNN for object detection based on small samples[J]. Defence Technology, 2020, 16(6): 1116-1129.

[18] 魏祥坡, 余旭初, 谭熊, 等. CNN和三维Gabor滤波器的高光谱图像分类[J]. 计算机辅助设计与图形学学报, 2020, 32(01): 90-98.

[19] Kovac I , Marak P . Finger vein recognition: utilization of adaptive gabor filters in the enhancement stage combined with sift/surf-based feature extraction[J]. Signal, Image and Video Processing, 2023, 17(3): 635-641.

[20] Imamura A, Arizymi N . Gabor filter incorporated CNN for compression[C]. 2021 36th International Conference on Image and Vision Computing New Zealand (IVCNZ). IEEE, 2021: 1-5.

[21] 蒋心璐, 陈天恩, 王聪, 等. 农业害虫检测的深度学习算法综述[J]. 计算机工程与应用, 2023, 59(06): 30-44.

[22] 周爱明, 马鹏鹏, 席天宇, 等. 基于深度学习的蝴蝶科级标本图像自动识别[J]. 昆虫学报, 2017(11): 1339-1348.

[23] 张洁. 基于图像识别的农作物害虫诊断技术研究与应用[D].合肥:中国科学技术大学, 2014.

[24] Sandler M, Howard A, Zhu M, et al. Mobilenetv2: Inverted residuals and linear bottlenecks[C]. Proceedings of the IEEE conference on computer vision and pattern recognition, 2018: 4510-4520.

[25] 李国进, 黄晓洁, 李修华, 等. 采用轻量级网络MobileNetV2的酿酒葡萄检测模型[J]. 农业工程学报, 2021, 37(17): 168-176.

[26] 刘张, 郭旭东, 李胜男. 基于MobileNetV2的小肠溃疡病灶图像识别[J]. 中国生物医学工程学报, 2024, 43(01): 70-79.

[27] 杨国亮, 李放, 朱晨, 等. 改进MobileNetV2网络在遥感影像场景分类中的应用[J]. 遥感信息, 2020, 35(01): 1-8.

[28] Su Z, Pietikainen M, Liu L . From Local Binary Patterns to Pixel Difference Networks for Efficient Visual Representation Learning[C]. Scandinavian Conference on Image Analysis. Cham: Springer Nature Switzerland, 2023: 138-155.

[29] Iqbal N, Mumtaz R, Shafi U, et al. Gray level co-occurrence matrix(GLCM) exture based crop classification using low altitude remote sensing platforms[J]. PeerJ Computer Science, 2021, 7: e536.

[30] Tang L, Ma S, Ma X, et al. Research on image matching of improved sift algorithm based on stability factor and feature descriptor simplification[J]. Applied Sciences, 2022, 12(17): 8448.

[31] Lee T S . Image representation using 2D Gabor wavel-ets[J]. IEEE Transactions on pattern analysis and machine intelligence, 1996, 18(10): 959-971.

[32] Ghiasi-Shirazi K . Learning 2D Gabor filters by infinite kernel learning regression[J]. Journal of Computational Mathematics and Data Science, 2021, 1: 100016.

[33] Howard A G, Zhu M, Chen B, et al. Mobilenets: Efficient convolutional neural networks for mobile vision applications[J]. arXiv preprint arXiv:1704.04861, 2017: 1-9.

[34] Savargiv M, Masoumi B, Keyvanpour M R . A new random forest algorithm based on learning automata[J]. Computational Intelligence and Neuroscience, 2021(1): 5572781.

[35] Tarzanagh D A, Li Y, Thrampoulidis C, et al. Transformers as support vector machines[J]. arXiv preprint arXiv:2308.16898, 2023: 1-58.

[36] Raghuvanshi S , Dhariwal S . The VGG16 Method Is a Powerful Tool for Detecting Brain Tumors Using Deep Learning Techniques[J]. Engineering Proceedings, 2023, 59(1): 46.

[37] Szegedy C , Vanhoucke V , Ioffe S , et al. Rethinking the inception architecture for computer vision[C]. Proceedings of the IEEE conference on computer vision and pattern recognition, 2016: 2818-2826.

[38] Zhang Q . A novel ResNet101 model based on dense dilated convolution for image classification[J]. SN Applied Sciences, 2022, 4: 1-13.

基金资助

云南省教育厅科学研究基金项目(2022J0493)