Spectral Open Set Recognition in Agriculture and Forestry Biological
Species Based on Fuzzy Rule Binary Classifier Combinations
HE Bao-xiong1, ZHAO Peng1*, LI Zhen-yu2
1. School of Computer Science and Software Engineering, Guangxi University of Science and Technology, Liuzhou 545006, China
2. School of Computer and Control Engineering, Northeast Forestry University, Harbin 150040, China
摘要： 开集分类识别要求分类器不仅能够“辨识”已知类别的测试样本，而且还要有效地“拒识”未知类别的测试样本；在光谱分析中有关的研究与应用相对较少。改进了Ishibuchi提出的经典的闭集框架下的模糊规则多类别分类器，将其应用于开集分类识别领域。首先，使用主成分分析法进行原始光谱曲线向量的光谱维度约简，降维至4维~6维的光谱特征向量。其次，将Ishibuchi提出的模糊规则多类别分类器简化为二元分类器版本，采用1-vs-1二元分类器进行分类处理，并且确定该测试样本在相应类别的得票。最后，将所有二元分类器的投票数进行统计，如果某个已知类别的得票数最高，并且该最高得票数大于预先确定的阈值τ，那么测试样本判决为该已知类别；否则就“拒识”为未知类别，从而实现了多类别的开集分类识别。在实验验证中，对于木材和芒果光谱数据集进行了分组的对比实验，结果表明，本方法优于其他的主流的开集分类识别，包括基于广义基本概率分配(generalized Basic probability assignment, GBPA）的改进的开集框架下的模糊规则多类别分类器；具有最好的评价指标F-Score, Kappa系数及总体识别率。此外，还针对芒果光谱数据集的对比实验进行了双尾McNemar's Test统计检验，进一步表明该方法相对于其他的开集分类识别方法来说，具有统计检验意义的优势。
关键词：开集分类识别；模糊规则分类器；二元分类器；光谱分析；统计检验
Abstract：Open set recognition requires that a classifier can not only identify testing samples from known classes but also reject those from unknown classes, which is rarely investigated in spectral analysis. In this article, we revise the conventional fuzzy rule multi-class classifier proposed by Ishibuchi for the closed-set scenario and apply it to open-set recognition. First, principal component analysis is used to reduce the spectral dimension of the original spectral curves, yielding 4- to 6-dimensional spectral feature vectors. Second, the fuzzy rule multi-class classifier proposed by Ishibuchi is simplified to a binary classifier, using a 1-vs-1 scheme to obtain a vote for each testing instance. Lastly, all votes from all binary classifiers are counted to determine the predicted class of the testing instance in the open-set scenario. If one known class gets the maximal vote and this vote is larger than a predetermined threshold τ, this testing instance is classified as this known class. Otherwise, it is rejected as an unknown class. The comparative experimental results across different groups of wood and mango spectral datasets indicate that our proposed scheme outperforms other state of the art open-set recognition schemes, such as the revised fuzzy rule multi-class classification based on generalized basic probability assignment, in the open-set scenario, with the best evaluation measures such as F-Score, Kappa coefficient, and overall recognition accuracy. Moreover, a dual-tailed McNemar's test is performed on the comparative experimental results from the mango spectral dataset to verify further that our proposed scheme is superior to other state of the art open-set recognition schemes.
Key words：Open set recognition; Fuzzy rule classifier; Binary classifier; Spectral analysis; Statistical test
何保雄，赵 鹏，李振宇. 基于模糊规则二元分类器组合的农林物种光谱开集分类识别研究[J]. 光谱学与光谱分析, 2025, 45(12): 3349-3357.
HE Bao-xiong, ZHAO Peng, LI Zhen-yu. Spectral Open Set Recognition in Agriculture and Forestry Biological
Species Based on Fuzzy Rule Binary Classifier Combinations. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2025, 45(12): 3349-3357.
[1] Geng C X, Huang S J, Chen S C. IEEE Transactions On Pattern Analysis and Machine Intelligence, 2021, 43(10): 3614.
[2] Scheirer W J, Rocha A D R, Sapkota A, et al. IEEE Transactions On Pattern Analysis and Machine Intelligence, 2013, 35(7): 1757.
[3] Scheirer W J, Jain L P, Boult T E. IEEE Transactions On Pattern Analysis and Machine Intelligence, 2014, 36(11): 2317.
[4] Zhang H, Patel V M. IEEE Transactions On Pattern Analysis and Machine Intelligence, 2017, 39(8): 1690.
[5] Bendale A, Boult T. 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, Hou C P, Lang Y, et al. Pattern Recognition, 2019, 85: 60.
[9] Hastie T, Tibshirani R. The Annals of Statistics, 1998, 26(2): 451.
[10] Dietterich T G, Bakiri G. Journal of Artificial Intelligence Research, 1995, 2(1): 263.
[11] Allwein E L, Shapire R E, Singer Y. Journal of Machine Learning Research, 2000, 1(1): 113.
[12] Garcia-Pedrajas N, Ortiz-Boyer D. IEEE Trans Pattern Analysis and Machine Intelligence, 2006, 28(6): 1001.
[13] Pujol O, Radeva P, Vitria J. IEEE Transactions On Pattern Analysis and Machine Intelligence, 2006, 28(6): 1007.
[14] Rocha A, Goldenstein S K. IEEE Transactions On Neural Networks and Learning Systems, 2014, 25(2): 289.
[15] LI Zhen-yu, ZHAO Peng, WANG Cheng-kun（李振宇，赵 鹏，王承琨）. Spectroscopy and Spectral Analysis（光谱学与光谱分析），2024，44（7）: 1868.
[16] Ishibuchi H, Nozaki K, Tanaka H. Fuzzy Sets and Systems, 1992, 52: 21.
[17] Ishibuchi H, Nakashima T. IEEE Transactions On Systems, Man and Cybernetics, Part B, 1999, 29(5): 601.
[18] Ishibuchi H, Nakashima T, Morisawa T. Fuzzy Sets and Systems, 1999, 103: 223.
[19] Bombardier V, Schmitt E. Engineering Applications of Artificial Intelligence, 2010, 23: 978.
[20] Anderson N T, Walsh K B, Flynn J R, et al. Postharvest Biology and Technology, 2021, 171: 111358.
[21] Tax D M J, Duin R P W. Machine Learning, 2004, 54(1): 45.
[22] Rodriguez A, Laio A. Science, 2014, 344(6191): 1492.
[23] Dietterich T G. Neural Computation, 1998, 10(7): 1895.
