Automatic used mobile phone color determination: Enhancing the used mobile phone recycling in China

https://doi.org/10.1016/j.resconrec.2022.106627Get rights and content

Highlights

  • By 2020, China will have a stock of 2 billion used mobile phones., but only 2% of these will have been recycled through formal channels.

  • Automatic color recognition of used and end-of-life mobile phones is essential to enable accurate and rapid recycling.

  • This study establishes a high-dimensional spatial color conversion deep convolutional neural network (HSCCNet) for mobile phone color recognition, which has a significant impact on the recycling of used mobile phones.

Abstract

Rapid development of telecommunication technology in China has led to a prosperous market of smart phones, as well as an increase number of used phones. Nevertheless, there are key factors affecting the used phone recycling, one of which is the phone color. To realize an accurate automatic color recognition of used phones to enhance the recycling process, a high-dimensional spatial color conversion deep convolutional neural network (HSCCNet) is proposed in this paper. First, we established a common dataset for the field of used electronic devices. Second, the phone color is converted to the high-dimensional space of hue, saturation and value (HSV), which generates richer expressions of color features and improves the model sensitivity. Finally, a deep convolutional structure for HSV features is designed, where color feature conversion are implemented, resulting in enhanced color feature expressions. Promising results are obtained through the comparison between the proposed HSCCNet and the state-of-the-art models.

Introduction

In recent years, the global mobile communication business has made great progress, especially the rapid development of China's mobile communication industry(Liu et al., 2019; Zeng et al., 2016). With the rapid turnover of mobile phone products, more and more used mobile phones are flowing into the recycling market, but the current status of used mobile phone recycling is not optimistic (Zhang et al., 2021). According to the October 2021 Domestic Mobile Phone Market Operation Analysis Report released by the China ICT Institute, the number of mobile phones eliminated was as high as 498 million, but the recycling rate was only 4.2%. Among them, 4% were recycled by traditional channels and 0.2% by internet channels. From 2014 to the present, the cumulative stock of used mobile phones in China has exceeded 2 billion units (Wang et al., 2022). If properly disposed of, it will bring huge emergency benefits and reduce environmental pollution (Wang et al., 2020). As a waste product with strong price timeliness, the recycling efficiency of used mobile phones seriously affects their recycling price (Hu et al., 2015). It is difficult to meet the requirements of real-time inspection using manual methods and to achieve accurate detection of surface defects (Baxter and Gram-Hanssen, 2016; Choi et al., 2010). In order to overcome the difficulties in detecting the complex and diverse surface defects, an automatic detection method for surface defects of used mobile phones is designed that can free up manpower and reduce costs, help mobile phone repair manufacturers to realize specific needs such as assessing and recording the surface damage of mobile phones, and improve the efficiency of manufacturers. However, the recognition accuracy of the method is low and cannot meet the high precision needs of enterprises (Mao et al., 2021; Zhang et al., 2021). The process of the recycling system for used mobile phones in China is shown in Fig. 1. Users give away their unused mobile phones to others for disposal or for recycling. Recycling methods are usually both online and offline. The recycled mobile phones are sent to the inspection center for processing, and the basic attributes are first identified, including defect identification, defect classification, model identification and color identification. The phones that can be sold second time are put into the recycling market, and the used mobile phones are processed for component recycling, precious metal extraction, etc. It is worth noting that the color of the mobile phone is an important recycling attribute in the recycling as well as secondary sales process, and with the various colors of mobile phones coming out, the color of the mobile phone has become one of the important indicators for the evaluation of the value of the mobile phone.

China has a large stock of used mobile phones, fast growth rate and low recycling rate. In the recycling process of used mobile phones, the color of mobile phones seriously affects the value of mobile phones, so a deep learning-based color recognition method for used mobile phones is proposed in the paper. Motivated by the above, this paper aims to propose a high-dimensional spatial color conversion deep convolutional neural network (HSCCNet) for mobile phone color recognition.

The proposed HSCCNet using the HSV color space as the input channel of the model. Compared with the traditional RGB color space, the HSV color space decouples the three components directly into the ranges of color, saturation and luminance, which are clearly and completely defined for the color space (Bargshady et al., 2020; Bora, 2017). Therefore, the model has a richer and more accurate color representation under HSV three channels. And targeting the HSV color features extracted from the network, the HSV color conversion layer is designed to effectively focus the extracted HSV color space features to the target-sensitive feature color regions, further enhancing the representation of color features.

Concretely, the contributions of this paper can be concluded in the following:

  • 1)

    We established a common dataset in the field of used electronic devices. The dataset is representative for used mobile phones and can be used for general deep learning model testing purpose, which is beneficial for the recycling industry.

  • 2)

    A richer feature expression of mobile phone colors is established and the model sensitivity is improved by converting the common color space to the color of the high-dimensional space of the hue, saturation, and value.

  • 3)

    A high-dimensional spatial color conversion deep convolutional neural network is designed. It effectively converts the extracted features into color sensitive region features and enhances color feature expression.

The rest of this paper is organized from the following aspects: Related work are discussed in  Section II. Feature analysis and data processing are given in Section III. The method of recognizing the colors of used mobile phones is introduced in Section IV. The experiments and analyzes the results are designed in Section V. Finally, conclusions are drawn in Section VI.

Section snippets

Related work

Currently, the color of mobile phones has become one of the important factors affecting the recycling of mobile phones, and many academics have launched many related research works on color recognition (Huang and Xu, 2019). A target image color recognition method based on RGB color features was proposed (Chen et al., 2014). This target color sample approach was created using the sector sampling method to extract the target image's color attributes and achieve a high recognition accuracy.

Data processing

The structure of the traditional RGB color space is a color representation method for the computer coding, which is significantly different from the color perception method of the human eyes. However, the physiological mechanism of the human eye color discrimination is used for reference by the HSV hexagonal cone color space model. This method separates the chromaticity and brightness in the color information, and can more clearly express the visual hue difference between colors. The hexagonal

Model architecture design based on HSCCNet

It is difficult to accurately recognize and classify colors of used mobile phones, a high-dimensional spatial color conversion deep convolutional neural network (HSCCNet) for mobile phone color recognition is proposed in the paper. The network structure is shown in Fig. 2, and it should be declared here that we only show the specific construction in a single block, and the actual network structure parameters are shown in Fig. A2–1, which consists of several groups of such blocks to form the

Database construction

During the used mobile phone data capture process, we build a representative dataset, where the used phone data were collected from used mobile phone companies in Huaqiang North Commercial Area, Beijing and Shanghai, which can represent the common used phone markets in China. We used mobile phones and digital cameras to capture color images of scrap mobile phones and collated the captured image data into a database of used mobile phone colors. The data were collected in July 2021 and were

Discussion

A report from the Ministry of Industry and Information Technology shows that China has a cumulative stock of more than 2 billion used mobile phones, with more than half a billion generated in 2020 alone. However, in contrast to the huge market for used mobile phones, only 2% of these phones are recycled through formal channels (Ministry of industry and information technology of the people's republic of China, 2021). The majority of used mobile phones still have use value and can be sold for a

Conclusion

In this paper, a color recognition method based on deep convolutional neural networks is proposed. The module based on multi-layer CNN networks is regarded as the basic feature extraction structure, and the adaptive optimization of the model can obtain higher mean average precision. This method is applicable to all major mobile phones on the market. Using the method proposed in this paper can accurately identify the color of used mobile phones, which has a facilitating effect on the recycling

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This work was supported by National Science Foundation of China under Grants 61890930–5, 61903010, 62021003 and 62125301, National Key Research and Development Project under Grant 2018YFC1900800–5 and 2018YFC1900804, Beijing Natural Science Foundation under Grant KZ202110005009, and Beijing Outstanding Young Scientist Program under Grant BJJWZYJH 01201910005020.

References (36)

  • M.K. Alsmadi et al.

    Fish recognition based on robust features

    J. Theor. Appl. Inf. Technol.

    (2010)
  • A.H. Ashtari et al.

    An iranian license plate recognition system based on color features

    IEEE Trans. Intelligent Transp. Syst.

    (2014)
  • Z.N. Binti et al.

    Fire recognition using RGB and ycbcr color space

    J. Eng. Appl. Sci.

    (2015)
  • D.J. Bora

    A novel approach for color image edge detection using multidirectional Sobel filter on HSV color space

    Int. J. Comput. Sci. Eng.

    (2017)
  • Y.-.H. Chang et al.

    Enhanced image captioning with color recognition using deep learning methods

    Appl. Sci.

    (2021)
  • P. Chen et al.

    Vehicle color recognition on urban road by feature context

    IEEE Trans. Intelligent Transp. Syst.

    (2014)
  • J.Y. Choi et al.

    Boosting color feature selection for color face recognition

    IEEE Trans. Image Process.

    (2010)
  • V.L.d.S. Freitas et al.

    Automatic shadow detection in aerial and terrestrial images

    Boletim de Ciências Geodésicas

    (2017)
  • Cited by (2)

    • Attention-Aware Based Differentiable Architecture Search Network for Waste Recognition

      2023, Proceedings of 13th IEEE International Conference on CYBER Technology in Automation, Control, and Intelligent Systems, CYBER 2023
    View full text