Tree Class Recognition in Open Set Based on an Improved Fuzzy
Reasoning Classifier
LI Zhen-yu1, ZHAO Peng1, 2*, WANG Cheng-kun3
1. College of Computer and Control Engineering, Northeast Forestry University, Harbin 150040, China
2. School of Computer Science and Software Engineering, Guangxi University of Science and Technology, Liuzhou 545006, China
3. School of Electronics and Information Engineering, Heilongjiang University of Science and Technology, Harbin 150022, China
摘要： 开集分类识别是近10多年来模式识别领域研究的热点，它能够识别训练集中已知类别的测试样本，同时还能够有效“拒识”未知类别的测试样本；这些未知类别样本不包含在训练集中。现有的开集分类识别算法主要是基于Support Vector Machine（SVM）和深度学习网络框架进行改进，并且主要应用在自然景物图像领域中；在光谱分析领域中还鲜有报道。将传统的闭集框架下的模糊推理分类器进行模型改进，提出了开集框架下的改进模糊推理分类器，并将其应用到木材树种近红外光谱分类识别中。首先，使用Flame-NIR近红外微型光谱仪采集木材样本横切面的近红外光谱曲线，采用Metric Learning算法进行光谱向量维度约简降维至4维（4D）。其次，改进闭集框架下的模糊推理分类器，根据模糊规则置信度和各维度隶属度概率的乘积构建Generalized Basic Probability Assignment（GBPA），再根据GBPA进行分类处理。在20个树种的具有不同的Openness指标下的近红外光谱数据集的分类识别对比实验表明，改进的开集模糊推理分类器(fuzzy reasoning classifier in an open set, FRCOS)优于现有的基于机器学习和深度学习的开集分类识别主流算法，具有较好的评价指标F-Score, Kappa系数及总体识别率。
关键词：开集分类识别；木材树种识别；模糊推理分类器；近红外光谱分析
Abstract：Open set recognition (OSR) has been investigated for approximately 10 years. It can recognize samples from the known classes in the training dataset, whereas it rejects samples from the unknown classes not included in the training dataset. The current OSR schemes are mainly based on Support Vector Machine (SVM) and deep learning neural networks. These OSR schemes are mainly used in natural scenery images and are rarely used in spectral analysis. In this paper, the classical fuzzy reasoning classifier in the closed set is improved with application to tree class spectral classification in the open set. First, a Flame-NIR spectrometer picks up the wood near-infrared (NIR) spectral curve. After metric learning processing, the spectral 4-dimensional (4D) feature vector is used as a classification feature. Second, the fuzzy reasoning classifier is improved for its use in an open set scenario. A new generalized basic probability assignment (GBPA) is used based on the confidence value of a fuzzy rule and the product of membership degree probability in each dimension. The comparison experimental results on wood NIR datasets with different “Openness” values indicate that our proposed scheme (Fuzzy Reasoning Classifier in an Open Set, FRCOS) outperforms the state of the art OSR schemes based on machine learning and deep learning with good performance evaluation measures.
Key words：Open set recognition; Tree class recognition; Fuzzy reasoning classifier; Near-infrared spectral analysis
李振宇，赵 鹏，王承琨. 改进模糊推理分类器进行木材树种近红外光谱开集分类识别研究[J]. 光谱学与光谱分析, 2024, 44(07): 1868-1876.
LI Zhen-yu, ZHAO Peng, WANG Cheng-kun. Tree Class Recognition in Open Set Based on an Improved Fuzzy
Reasoning Classifier. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2024, 44(07): 1868-1876.
[1] Scheirer W J, Rocha A D R, Sapkota A, et al. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013，35(7)：1757.
[2] Scheirer W J, Jain L P, Boult T E. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014，36(11)：2317.
[3] Jain L P, Scheirer W J, Boult T E. Multi-Class Open Set Recognition Using Probability of Inclusion. In Computer Vision—ECCV 2014. Springer, 2014. 393.
[4] Zhang H，Patel V. IEEE Trans Pattern Analysis and Machine Intelligence, 2017，39(8)：1690.
[5] Bendale A，Boult T E. Towards Open World Recognition. Proceedings of IEEE Computer Vision and Pattern Recognition (CVPR), 2015. 1893.
[6] Junior P R M, Souza R M D, Werneck R D O, et al. Machine Learning, 2017，106(3)：359.
[7] Bendale A，Boult T E. Towards Open Set Deep Networks. Proceedings of IEEE CVPR, 2016. 1563.
[8] Yang Y, Chun P H, Yue L, et al. Pattern Recognition, 2019, 85：60.
[9] Ge Z Y, Demyanov S, Chen Z, et al. arXiv, 2017, 1707: 07418.
[10] Geng C X, Huang S J, Chen S C. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021，43(10)：3614.
[11] Deng J, Dong W, Socher R, et al. ImageNet: A Large-scale Hierarchical Image Database. Proceedings of 2009 IEEE Conference on Computer Vision and Pattern Recognition, 2009. 248.
[12] Russakovsky O, Deng J, Su H, et al. International Journal of Computer Vision, 2015，115(3)：211.
[13] Lavine B K, Davidson C E, Moores A J, et al. Applied Spectroscopy, 2001，55(8)：960.
[14] WANG Yuan, ZHE Shuai, ZHOU Nan, et al(王 远，折 帅，周 南，等). Spectroscopy and Spectral Analysis(光谱学与光谱分析)，2019，39（9）: 2719.
[15] Ma T, Inagaki T, Tsuchikawa S. Holzforschung, 2021，75(5): 419.
[16] Bombardier V, Mazaud C, Lhoste P, et al. Computers in Industry, 2007，58：355.
[17] Bombardier V, Schmitt E. Engineering Applications of Artificial Intelligence, 2010，23：978.
[18] Bombardier V, Schmitt E, Charpentier P. Measurement, 2009，42(2)：189.
[19] Mensink T, Verbeek J, Perronnin F, et al. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013，35(11)：2624.
[20] Nakashima T, Schaefer G, Yokota Y, et al. Fuzzy Sets and Systems, 2007，158：284.
[21] Ishibuchi H, Nakashima T. IEEE Transactions on Industrial Electronics, 1999，46(6)：1057.
[22] Tax D M J, Duin R P W. Machine Learning, 2004，54(1)：45.
[23] Rodriguez A, Alessandro L. Science, 2014，344(6191)：1492.
