基于RF-LCE-BiLSTM-Attention-AMSSA模型的京剧二分类

doi:10.3969/j.issn.1006-2475.2024.11.002

摘要/Abstract

摘要： 为了提升大数据背景下的京剧分类精度，促进国粹传播，本文在RF-BiLSTM-Attention模型基础上利用偏差惩罚交叉熵（LCE）损失函数防止过拟合，并融合自适应麻雀优化算法（AMSSA）提出RF-LCE-BiLSTM-Attention-AMSSA模型对京剧与其他音乐进行二分类。模型首先将音频文件转化为由特征构成的向量，再结合L2正则化损失和交叉熵损失作为模型的偏差惩罚交叉熵损失函数，通过神经网络进行分类训练；然后利用AMSSA优化模型超参数，将最优超参数代入模型进行京剧二分类。以来自大众音乐平台和GTZAN数据集的1500首音乐进行京剧二分类实验，对比RF-LCE-BiLSTM-Attention-AMSSA模型与RNN、LSTM、BiLSTM等10种模型的分类准确率，并对LCE损失函数与AMSSA进行消融实验。实验结果表明：RF-LCE-BiLSTM-Attention-AMSSA模型的分类准确率为89.95%，比RF-BiLSTM-Attention提高了0.95个百分点；比RF-LCE-BiLSTM-Attention提高了0.28个百分点。

关键词: 京剧, 双向长短时记忆网络, 交叉熵, 自适应麻雀优化算法, 二分类

Abstract: In order to improve the classification accuracy of Beijing Opera in the era of big data and promote the dissemination of national essence， this article uses the deviation penalty cross entropy loss function on the basis of RF-BiLSTM-Attention to prevent overfitting and integrates the Adaptive Multi-Swarm Sparrow Search Algorithm （AMSSA） to propose the RF-LCE-BiLSTM-Attention-AMSSA model for binary classification of Beijing Opera and other music. The model first converts audio files into feature vectors， and then combines L2 regularization loss and Cross Entropy loss （LCE） as the deviation penalty cross entropy loss function of the model， which is trained through neural networks for classification. After that， the AMSSA is adopted to optimize the hyperparameters， and the optimal hyperparameters are applied for the binary classification of Beijing Opera. A Beijing Opera binary classification experiment was conducted on 1500 pieces of music， which come from the popular music platform and GTZAN dataset， to compare the classification accuracy of RF-LCE-BiLSTM-Attention-AMSSA model with 11 models such as RNN， LSTM， and BiLSTM， and to compare the impact of LCE loss function and AMSSA on the model. The results show that the classification accuracy of RF-LCE-BiLSTM-Attention-AMSSA model is 89.95%， which is 0.95 percentage points higher than RF-BiLSTM-Attention， and 0.28 percentage points higher than RF-LCE-BiLSTM-Attention.

Key words: , Beijing Opera, BiLSTM, cross entropy, AMSSA, binary classification

中图分类号:

TP183

龚谊承1, 2, 刘青1, 2. 基于RF-LCE-BiLSTM-Attention-AMSSA模型的京剧二分类[J]. 计算机与现代化, 2024, 0(11): 7-12.

GONG Yicheng1, 2, LIU Qing1, 2. Beijing Opera Binary Classification Based on RF-LCE-BiLSTM-Attention-AMSSA Model [J]. Computer and Modernization, 2024, 0(11): 7-12.

参考文献

［1］王文照. 北京京剧票房运营调研报告：以“北京新荣春青年京剧社”为例［D］. 北京：中国艺术研究院. 2021.
［2］袁思源. X大学附中“京剧社”发展现状研讨［D］. 重庆：西南大学， 2018.
［3］ MUTANU L， GOHIL J， GUPTA K， et al. A review of automated bioacoustics and general acoustics classification research［J］. Sensors， 2022，22（21）. DOI: 10.3390/s22218361.
［4］ SABHA A， SELWAL A. Data-driven enabled approaches for criteria-based video summarization： A comprehensive survey， taxonomy， and future directions［J］. Multimedia Tools and Applications， 2023，82（21）：32635-32709.
［5］ HAMA SAEED M. Improved speech emotion classification using deep neural network［J］. Circuits， Systems， and Signal Processing， 2023，42（12）：7357-7376.
［6］ Rishabh， KUMAR D. Multi spectral feature extraction to improve lung sound classification using CNN［C］// 2023 10th International Conference on Signal Processing and Integrated Networks （SPIN）. IEEE， 2023：186-191.
［7］ DORAISAMY S， GOLZARI S， NAROWI N M， et al. A study on feature selection and classification techniques for automatic genre classification of traditional Malay music［C］// 9th International Conference on Music Information Retrieval（ISMIR 2008）. ISMIR， 2008：331-336.
［8］ PEETERS G. Spectral and temporal periodicity representations of rhythm for the automatic classification of music audio signal［J］. IEEE Transactions on Audio， Speech， and Language Processing， 2011，19（5）：1242-1252.
［9］ ABEßER J. A review of deep learning based methods for acoustic scene classification［J］. Applied Sciences， 2020，10（6）. DOI： 10.3390/app10062020.
［10］ ABAYOMI-ALLI O O， DAMAŠEVIČIUS R， QAZI A， et al. Data augmentation and deep learning methods in sound classification： A systematic review［J］. Electronics， 2022，11（22）. DOI: 10.3390/electronics11223795.
［11］ ISMAIL FAWAZ H， FORESTIER G， WEBER J， et al. Deep learning for time series classification： A review［J］. Data Mining and Knowledge Discovery， 2019，33（4）：917-963.
［12］ WEI S Y， ZOU S， LIAO F F， et al. A comparison on data augmentation methods based on deep learning for audio classification［C］// 2019 2nd International Conference on Computer Information Science and Artificial Intelligence （CISAI 2019）. Journal of Physics： Conference Series. IOP Publishing， 2020，1453. DOI: 10.1088/1742-6596/1453/1/012085.
［13］ ZINEMANAS P， ROCAMORA M， MIRON M， et al. An interpretable deep learning model for automatic sound classification［J］. Electronics， 2021，10（7）. DOI: 10.3390/electronics10070850.
［14］ QAMHAN M A， ALTAHERI H， MEFTAH A H， et al. Digital audio forensics： Microphone and environment classification using deep learning［J］. IEEE Access， 2021，9：62719-62733.
［15］ LIU X Y， LU H L， YUAN J B， et al. CAT： Causal audio transformer for audio classification［C］// IEEE International Conference on Acoustics， Speech and Signal Processing（ICASSP 2023）. IEEE， 2023. DOI: 10.1109/ICASSP49357.2023.10096787.
［16］ CHEN K， DU X J， ZHU B L， et al. HTS-AT： A hierarchical token-semantic audio transformer for sound classification and detection［C］// IEEE International Conference on Acoustics， Speech and Signal Processing（ICASSP 2022）. IEEE， 2022：646-650.
［17］ ZHU W T， OMAR M. Multiscale audio spectrogram transformer for efficient audio classification［C］// IEEE International Conference on Acoustics， Speech and Signal Processing（ICASSP 2023）. IEEE， 2023. DOI: 10.1109/ICASSP49357.2023.10096513.
［18］龚谊承，刘青，肖浩逸. 基于RF-BiLSTM-Attention音乐分类方法的京剧二分类仿真［J］. 广西科学院学报， 2023，39（3）：322-330.
［19］ HO Y S， WOOKEY S. The real-world-weight cross-entropy loss function： Modeling the costs of mislabeling［J］. IEEE Access， 2019，8：4806-4813.
［20］ BÜHLMANN P， YU B. Boosting with the L2 loss： Regression and classification［J］. Journal of the American Statistical Association， 2003，98（462）：324-339.
［21］唐延强，李成海，宋亚飞，等. 自适应变异麻雀搜索优化算法［J］. 北京航空航天大学学报， 2023，49（3）：681-692.
［22］ CHEN R C， DEWI C， HUANG S W， et al. Selecting critical features for data classification based on machine learning methods［J］. Journal of Big Data， 2020，7. DOI: 10.1186/s40537-020-00327-4.
［23］ DATTA A， NG K F， BALAKRISHNAN D， et al. A data reduction and compression description for high throughput time-resolved electron microscopy［J］. Nature Communications， 2021，12. DOI: 10.1038/s41467-020-20694-z.
［24］ LI P H， FU T J， MA W Y. Why attention? Analyze BiLSTM deficiency and its remedies in the case of NER［C］// Proceedings of the 2020 AAAI Conference on Artificial Intelligence. AAAI， 2020，34（5）：8236-8244.
［25］ YIN X C， LIU Z G， LIU D Y， et al. A novel CNN-based Bi-LSTM parallel model with attention mechanism for human activity recognition with noisy data［J］. Scientific Reports， 2022，12. DOI: 10.1038/s41598-022-11880-8.
［26］黄庆康，宋恺涛，陆建峰. 应用于不平衡多分类问题的损失平衡函数［J］. 智能系统学报， 2019，14（5）：953-958.
［27］ DE LIMA R P， VAHEDI B， KARIMZADEH M. Comparison of cross-entropy， dice， and focal loss for sea ice type segmentation［C］// IEEE International Geoscience and Remote Sensing Symposium（IGARSS 2023）. IEEE， 2023：145-148.
［28］袁瑛. 基于正则化的多核学习方法及应用［D］. 广州：华南理工大学， 2016.
［29］耿志强，吴启明，江国华，等. XRT智能预选技术在有色金属矿山中的应用进展［J］. 铜业工程，2023（5）：119-126.
［30］ LAN G M， HOU C P， YI D Y. Robust feature selection via simultaneous capped ℓ2-norm and ℓ2， 1-norm minimization［C］// 2016 IEEE International Conference on Big Data Analysis （ICBDA 2016）. IEEE， 2016. DOI: 10.1109/IGARSS521008.2023.10282060.
［31］ GOUTTE C， GAUSSIER E. A probabilistic interpretation of precision， recall and F-score， with implication for evaluation［C］// 2005 European Conference on Information Retrieval. Springer， 2005：345-359.

[1]	徐涯昕, 何泽恩, 徐绪堪. 基于CNN-BiLSTM网络的数控机床故障文本自动分类[J]. 计算机与现代化, 2023, 0(04): 7-14.
[2]	盛江岸, 陈淑荣. 融合双重注意力机制的戴口罩人脸识别方法[J]. 计算机与现代化, 2023, 0(02): 72-77.
[3]	魏丽君, 吴海波, 章若冰. 一种基于信息测度的多属性决策方法[J]. 计算机与现代化, 2022, 0(06): 27-31.
[4]	王运乾, 王以松, 陈攀峰, 邹龙. 融合注意力机制的药用植物文本命名实体识别[J]. 计算机与现代化, 2021, 0(11): 100-105.
[5]	徐碧赢 1,孙慧 2. 基于交叉熵算法的NC-OFDM系统导频设计[J]. 计算机与现代化, 2017, 0(10): 62-65.