Optics and Precision Engineering, Volume. 31, Issue 2, 246(2023)

TCS-YOLO model for global oil storage tank inspection

Xiang LI1...2, Rigen TE1,2,*, Feng YI1,2 and Guocheng XU3 |Show fewer author(s)
Author Affiliations
  • 1Chang Guang Satellite Technology CO.,LTD., Changchun30000, China
  • 2Main Laboratory of Satellite Remote Sensing Technology of Jilin Province, Changchun130000, China
  • 3College of Materials Science and Engineering,Jilin University, Changchun10000, China
  • show less
    Figures & Tables(21)
    Schematic diagram of each structure of the backbone layer of YOLOv5
    Variation of γ with target frame size 1 600(40×40)
    Ground-truth and multiple predicted bounding boxes for the same tank object
    TCS-YOLO network structure
    Global oil storage tank dataset distribution
    Dataset diversity diagram
    Annotated oil storage tank dataset
    Experimental results of different object detection models
    Schematic diagram of oil storage tank target recognition by TCS-YOLO model
    • Table 1. Impact of C3TR layer replacement position on model performance

      View table
      View in Article

      Table 1. Impact of C3TR layer replacement position on model performance

      AlgorithmGFLOPsmAP0.5/%mAP0.5∶0.95/%
      YOLOv5107.899.289.8

      YOLOv5+C3TR

      (position=1)

      107.8--

      YOLOv5+C3TR

      (position=4)

      104.999.391.1

      YOLOv5+C3TR

      (position=6,7,8)

      99.099.189.9

      YOLOv5+C3TR

      (position=4,5,6,7,8)

      93.199.189.4
    • Table 2. Effects of LN on C3TR

      View table
      View in Article

      Table 2. Effects of LN on C3TR

      ModelGFLOPsmAP0.5/%mAP0.5∶0.95/%
      YOLOv5107.899.289.8

      YOLOv5+C3TR

      (with LN)

      107.899.189.3

      YOLOv5+C3TR

      (without LN)

      104.999.391.1
    • Table 3. Effect of CBAM layer replacement position on model performance

      View table
      View in Article

      Table 3. Effect of CBAM layer replacement position on model performance

      AlgorithmGFLOPsmAP0.5/%mAP0.5∶0.95/%
      YOLOv5107.899.289.8

      YOLOv5+CBAM

      (position=1)

      107.899.390.8

      YOLOv5+CBAM

      (position=4)

      108.099.290.7

      YOLOv5+CBAM

      (position=1,4)

      108.099.290.7

      YOLOv5+CBAM

      (position=1,2,3,4)

      108.199.290.5

      YOLOv5+CBAM

      (position=5,6,7,8)

      108.199.289.9

      YOLOv5+CBAM

      (position=1,2,3,4,5,6,7,8)

      108.199.290.6
    • Table 4. Different IoU methods and results

      View table
      View in Article

      Table 4. Different IoU methods and results

      IoU Method(A,G)(B,G)(C,G)(D,G)
      IoU0.30.30.60.6
      GIoU0.30.30.60.6
      DIoU0.30.30.6240.624
      CIoU0.30.30.6240.624
      SIoU0.3060.310.7210.862
    • Table 5. Influence of different IoU methods of localization loss functions on experimental results

      View table
      View in Article

      Table 5. Influence of different IoU methods of localization loss functions on experimental results

      Algorithm

      mAP0.5

      /%

      mAP0.5∶0.95

      /%

      YOLOv5+IoU loss99.189.6
      YOLOv5+GIoU loss99.289.8
      YOLOv5+DIoU loss99.289.8
      YOLOv5+CIoU loss(default)99.289.8
      YOLOv5+SIoU loss99.290.5
    • Table 6. Global oil storage tank dataset data source

      View table
      View in Article

      Table 6. Global oil storage tank dataset data source

      ContinentCountry/CityCompanyNumber of images

      Africa

      Algiers, AlgeriaSonatrach4
      Tripoli, LibyaNation Oil Corporation1
      Cape Town, South AfricaPetroSA1
      AsiaHong Kong, ChinaShell1
      Ningbo, ChinaSinopec1
      Xingang, ChinaChina National Petroleum Corporation2
      Tehran, IranNational Iranian Oil Company1
      Kawasaki, JapanShowa Shell Sekiyu5
      Osaka, JapanEneos4
      Doha, QatarQatar Petroleum1
      Jeddah, Saudi ArabiaSaudi Aramco1
      Singapore, SingaporeSingapore Petroleum Company Limited3
      Abu Dhabi, UAEAbu Dhabi National Oil Company6
    • Table 6. Global oil storage tank dataset data source

      View table
      View in Article

      Table 6. Global oil storage tank dataset data source

      ContinentCountry/CityCompanyNumber of images
      EuropeDen Haag, NetherlandRoyal Dutch Shell2
      North AmericaMexico City, MexicoPemex1
      San Francisco, USChevron Corporation1
      Houston, USExxon Mobil2
      OceaniaBrisbane, AustraliaSenex1
      Melbourne, AustraliaOil Company of Australia2
      South AmericaRiode Janeiro, BrazilPetrobras3
      Total--43
    • Table 7. Data set image parameter comparison

      View table
      View in Article

      Table 7. Data set image parameter comparison

      ImageFilming timeSolar azimuthSatellite azimuthCloudiness
      Abu Dhabi-A2021-05-10 21:39:2999.039 8278.6291%
      Abu Dhabi-B2020-05-17 23:33:5498.129 2100.9487%
    • Table 8. TCS-YOLO model training hyperparameter settings

      View table
      View in Article

      Table 8. TCS-YOLO model training hyperparameter settings

      HyperparametersValue
      Input resolution416
      Initial learning Rate0.01
      Training epochs300
      Momentum0.937
      Weight decay0.000 5
    • Table 9. Batch-size value settings for different models

      View table
      View in Article

      Table 9. Batch-size value settings for different models

      ModelsBatch-size
      TCS-YOLO-n512
      TCS-YOLO-s256
      TCS-YOLO-m128
      TCS-YOLO-l64
      TCS-YOLO-x32
    • Table 10. Experimental results of different object detection models

      View table
      View in Article

      Table 10. Experimental results of different object detection models

      AlgorithmmAP0.5/%mAP0.5∶0.95/%Inference time/msParameters/MGFLOPs
      YOLOv3-tiny96.477.50.533.112.9
      YOLOv398.385.74.0243.9141.5
      YOLOv3-spp98.787.54.0238.6155.6
      YOLOv497.384.73.7234.6154.7
      YOLOv4-pacsp-mish-s97.884.21.330.720.4
      YOLOv4-pacsp-mish98.186.04.3200.3119.3
      YOLOv4-pacsp-mish-x98.286.86.0378.5220.8
      Swin_Transformer_mask_rcnn_tiny94.474.21.9326.65540.5
      Swin_Transformer_faster_rcnn95.379.04.3158.434.0
      Swin_Transformer_mask_rcnn_small96.882.05.6408.0555.7
      YOLOv5n99.084.80.36.74.2
      YOLOv5s99.088.40.826.916.3
      YOLOv5m99.189.71.980.350.3
      YOLOv5l99.289.83.1177.8114.1
      YOLOv5x99.289.95.3332.7217.1
      TCS-YOLO-n99.085.00.46.74.2
      TCS-YOLO-s99.389.40.926.815.7
      TCS-YOLO-m99.391.12.179.147.0
      TCS-YOLO-l99.491.33.4175.8111.2
      TCS-YOLO-x99.491.65.6328.0210.9
    • Table 11. Impact of different optimization methods on model performance

      View table
      View in Article

      Table 11. Impact of different optimization methods on model performance

      AlgorithmSIoUCBAMC3TRmAP0.5/%mAP0.5∶0.95/%
      YOLOv599.289.8
      YOLOv5+SIoU loss99.290.5
      YOLOv5+CBAM99.390.8
      YOLOv5+C3TR99.391.1
      YOLOv5+SIoU loss+CBAM99.391.2
      YOLOv5+SIoU loss+C3TR99.491.2
      YOLOv5+CBAM+C3TR99.491.2
      TCS-YOLO(YOLOv5l+C3TR+CBAM+SIoU loss)99.491.3
    Tools

    Get Citation

    Copy Citation Text

    Xiang LI, Rigen TE, Feng YI, Guocheng XU. TCS-YOLO model for global oil storage tank inspection[J]. Optics and Precision Engineering, 2023, 31(2): 246

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Information Sciences

    Received: Jul. 15, 2022

    Accepted: --

    Published Online: Feb. 9, 2023

    The Author Email: TE Rigen (terigen@jl1.cn)

    DOI:10.37188/OPE.20233102.0246

    Topics