• Acta Optica Sinica
  • Vol. 40, Issue 24, 2411001 (2020)
Wenhao Lai, Mengran Zhou*, Feng Hu, Kai Bian, and Hongping Song
Author Affiliations
  • College of Electrical and Information Engineering, Anhui University of Science and Technology, Huainan, Anhui 232000, China
  • show less
    DOI: 10.3788/AOS202040.2411001 Cite this Article Set citation alerts
    Wenhao Lai, Mengran Zhou, Feng Hu, Kai Bian, Hongping Song. Coal Gangue Detection Based on Multi-Spectral Imaging and Improved YOLO v4[J]. Acta Optica Sinica, 2020, 40(24): 2411001 Copy Citation Text show less
    System of multi-spectral data acquisition
    Fig. 1. System of multi-spectral data acquisition
    Multi-spectral image of coal and coal gangue
    Fig. 2. Multi-spectral image of coal and coal gangue
    Structures of YOLO and YOLO v3 models
    Fig. 3. Structures of YOLO and YOLO v3 models
    Schematic diagram of YOLO v4.1
    Fig. 4. Schematic diagram of YOLO v4.1
    Average recognition accuracy in different bands. (a) Two classes of coal and coal gangue; (b) three classes of coal, coal gangue, and mix
    Fig. 5. Average recognition accuracy in different bands. (a) Two classes of coal and coal gangue; (b) three classes of coal, coal gangue, and mix
    Confusion matrix of correlation coefficients
    Fig. 6. Confusion matrix of correlation coefficients
    Test results. (a) Coal416; (b) coal512; (c) coal608; (d) coal408; (e) coal gangue416; (f) coal gangue512; (g) coal gangue608; (h) coal gangue408
    Fig. 7. Test results. (a) Coal416; (b) coal512; (c) coal608; (d) coal408; (e) coal gangue416; (f) coal gangue512; (g) coal gangue608; (h) coal gangue408
    Detection result of YOLO v4.1
    Fig. 8. Detection result of YOLO v4.1
    SampleNumber of lumpsTotal
    123
    Coal1757550300
    Coal gangue1757550300
    Mix-100150250
    Table 1. Data information of coal and coal gangue
    AlgorithmCoal and coal gangueCoal, coal gangue, and mix
    Maximum average accuracy /%BandMaximum average accuracy /%Band
    RF97.86,7,1087.09
    CART89.01370.89
    AdaBoost98.01277.32,6
    Table 2. Recognition results of AdaBoost, RF, and CART
    Band971168101321
    Average accuracy87.086.786.786.486.486.486.486.4
    Table 3. [in Chinese]
    Detection modelInput resolution /(pixel×pixel)Average precisionmAP /%Test time /s
    Coal /%Coal gangue /%
    YOLO v4.1408×40898.7397.7898.264.18
    416×41697.4793.4995.483.43
    YOLO v4512×51298.6193.8796.244.33
    608×60898.5997.0497.816.07
    Table 4. Test results of each detection model
    SampleDetection result [label:score (xlt, ylt), (xrb, yrb)]
    LabelScore(xlt,ylt),(xrb,yrb)LabelScore(xlt,ylt),(xrb,yrb)LabelScore(xlt,ylt),(xrb,yrb)
    1c0.97(87,124),(173,254)--
    2c1.00(8,200),(109,275)c1.00(92,221),(184,297)-
    3c1.00(34,119),(130,209)c0.99(42,217),(173,319)c0.91(120,104),(216,223)
    4g0.58(90,145),(216,261)--
    5g1.00(43,242),(152,326)g0.98(51,209),(149,266)-
    6g0.84(7,178),(162,320)g0.81(79,152),(164,255)-
    7g1.00(6,110),(141,205)c0.72(44,205),(173,318)-
    8g0.96(29,100),(185,197)c0.99(41,202),(126,320)-
    9g0.98(106,134),(216,258)c0.99(26,84),(128,196)c0.99(29,212),(124,326)
    Table 5. Coordinate and score of bounding box
    Wenhao Lai, Mengran Zhou, Feng Hu, Kai Bian, Hongping Song. Coal Gangue Detection Based on Multi-Spectral Imaging and Improved YOLO v4[J]. Acta Optica Sinica, 2020, 40(24): 2411001
    Download Citation