Twelve-layer deep convolutional neural network with stochastic pooling for tea category classification on GPU platform
- 135 Downloads
Automatic tea-category identification is an important topic in factories and supermarkets. Traditional methods need to extract features from tea images manually, which may not be optimal for tea images classification. To avoid the time consuming efforts of handcrafted features extraction, this study proposed a new method combining convolutional neural network (CNN) with stochastic pooling. We collected 900 tea images of Oolong, green, and black teas, with 300 images for each category. The data augmentation method was used over the training set. We employed stochastic gradient descent with momentum (SGDM) to train the CNN. The experiments showed that a 12-layer CNN gives a good result. The sensitivities of Oolong, green, and black tea are 99.5%, 97.5%, and 98.0%, respectively. The overall accuracy of all three-tea categories is 98.33%. The stochastic pooling gives better results than maximum pooling and average pooling. The optimal number of convolutional layer for this task is 5. In addition, GPU has a 175× acceleration in training set and a 122× acceleration in test set, compared to CPU platform.
KeywordsConvolutional neural network Stochastic pooling Data augmentation Tea category classification Stochastic gradient descent with momentum
This paper is supported by Open fund of Key Laboratory of Guangxi High Schools Complex System and Computational Intelligence (2016CSCI01), Natural Science Foundation of China (61502254, 61602250), Program of Natural Science Research of Jiangsu Higher Education Institutions (15KJB470010, 16KJB520025), Natural Science Foundation of Jiangsu Province (BK20150983).
Compliance with ethical standards
Conflict of interest
There is no conflict of interest regarding the submission of this paper.
- 2.Barushka A, Hajek P (2016) Spam Filtering Using Regularized Neural Networks with Rectified Linear Units. In: 15th International Conference of the Italian Association for Artificial Intelligence (AIIA). Springer Int Publishing Ag, Genova, p 65–75Google Scholar
- 12.Gorriz JM, Ramírez J (2016) Wavelet entropy and directed acyclic graph support vector machine for detection of patients with unilateral hearing loss in MRI scanning. Front Comput Neurosci 10:160Google Scholar
- 13.Gummeson A et al (2017) Automatic Gleason grading of HE stained microscopic prostate images using deep convolutional neural networks. In: Medical Imaging 2017, Digital Pathology, 2017, 10140. https://doi.org/10.1117/12.2253620
- 18.Kabani A, El-Sakka MR (2016) Object Detection and Localization Using Deep Convolutional Networks with Softmax Activation and Multi-class Log Loss. In: 13th International Conference on Image Analysis and Recognition in Memory of Mohamed Kamel (ICIAR). Springer Int Publishing Ag, Povoa de Varzim, p 358–366Google Scholar
- 19.Li XL, Zhang YY, He Y (2017) Study on Detection of Talcum Powder in Green Tea Based on Fourier Transform Infrared (FTIR) Transmission Spectroscopy. Spectrosc Spectr Anal 37(4):1081–1085Google Scholar
- 21.Lu S, Lu Z (2016) A pathological brain detection system based on kernel based ELM. Multimed Tools Appl. https://doi.org/10.1007/s11042-016-3559-z
- 22.Martinez-Pabon F et al (2016) Recommending Ads from Trustworthy Relationships in Pervasive Environments. Mob Inf Syst 2016:8593173Google Scholar
- 26.Sun M et al (2016) Max-pooling loss training of long short-term memory networks for small-footprint keyword spotting. In: IEEE workshop on spoken language technology (SLT). IEEE, San Diego, p 474–480Google Scholar
- 29.Wu X (2016) Tea category identification based on optimal wavelet entropy and weighted k-Nearest Neighbors algorithm. Multimed Tools Appl. https://doi.org/10.1007/s11042-016-3931-z
- 30.Yang J (2015) Identification of green, Oolong and black teas in China via wavelet packet entropy and fuzzy support vector machine. Entropy 17(10):6663–6682Google Scholar
- 32.Zeiler M, Fergus R (2013) Stochastic pooling for regularization of deep convolutional neural networks. In Proceedings of the International Conference on Learning Representation (ICLR)Google Scholar
- 35.Zhu SG, Du JP (2014) Visual Tracking Using Max-Average Pooling and Weight-Selection Strategy. J Appl Math 2014:828907Google Scholar