Optics and Precision Engineering, Volume. 32, Issue 6, 868(2024)

Review of defect detection algorithms for solar cells based on machine vision

Yuqi LIU and Yiquan WU*
Author Affiliations
  • College of Electronic and Information Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing211106, China
  • show less
    Figures & Tables(22)
    Diagram of the entire process of crystalline silicon solar cell production
    Surface images of solar cells obtained with different imaging methods
    Framework of surface defect detection algorithm of solar cells based on traditional machine vision
    Framework of traditional machine vision solar cell defect detection algorithms
    Textural differences in edge gradients between cracks, broken grid defects and the background
    Diagram of the journey of development of deep learning networks
    Network framework diagram proposed in [61]
    Network framework diagram proposed in [41]
    Example of common datasets
    Confusion matrix
    Visualization of segmentation evaluation metrics
    • Table 1. Common defects of solar cells

      View table
      View in Article

      Table 1. Common defects of solar cells

      Defect categoryDefect nameImpact on cell performance

      Visual

      characteristics

      CausesSampling
      Shape defectsCracksResulting in the interruption of the current path, reducing the conversion efficiency of the cellTiny hidden cracks or larger cracksWelding or cutting errors, production process errors, impacts
      Broken gateBroken grid part can not collect the current, resulting in lower cell conversion efficiencyInterrupted or broken conductive grid lines
      Missing corners,BrokenMay cause damage to the grid line, affecting the module power factorMissing or redundant shape
      Color defectsSpots、Black spotsMay reduce the absorption of light by the solar cell, lowering the photovoltaic conversion efficiencyAbnormal or uneven surface colorationUneven chemical reaction during coating, defective raw silicon material
      Black flakesFailure of the black part of the solar cell to provide power may cause thermal breakdown, resulting in lower output power of the whole module
      Texture defects

      Water marks,Fingerprints,

      Wheel marks

      Resulting in uneven reflection or absorption of light from the surface and a decrease in photoelectric conversion efficiency in some areasAreas of abnormal brightness, speckles or fingerprints, wheel marks on the surfaceImproper manual operation, excessive machine pressure, untidy production environment
    • Table 2. Advantages and disadvantages of specific algorithms based on traditional machine vision

      View table
      View in Article

      Table 2. Advantages and disadvantages of specific algorithms based on traditional machine vision

      MethodAdvantagesLimitations
      Improved anisotropic diffusion filter24Accurate segmentation of hidden cracksOnly applicable for microcrack detection
      Mean shift technique25Significant detection of fingerprint and contamination defectsMethod lacks generality
      RBF+ SVM34

      Detection of multiple defects, strong

      interference resistance

      Does not meet real-time production inspection

      requirements

      KAZE+SVM35Low resource consumption, relatively short processing timeLower accuracy on polycrystalline units
      DBF+SVM41

      Good performance on imbalanced

      datasets, fast detection speed

      Limited labeled image data in the dataset
      Uniform distribution clustering45Improved detection accuracy for various types of crack defectsLimited to small datasets, has limitations on the size of the inspection images
      CPICS-LBP+SVM47

      Extraction of defect features in the

      presence of uneven background

      Only considers the average of single-layer adjacent pixel information
      Visual attention mechanism51Accurate localization of defect areas by the model

      Does not meet real-time monitoring production

      requirements

      PSO+SVM52

      Strong interference resistance,

      high generality

      Limited samples in the dataset
      ICA58Relatively high detection accuracy

      Unable to distinguish between grid breakage and

      microcracks

    • Table 3. Comparison of various algorithms based on traditional machine vision

      View table
      View in Article

      Table 3. Comparison of various algorithms based on traditional machine vision

      AlgorithmAdvantagesDisadvantagesApplicable image types
      Image domain analysisGradient featuresRelatively simple algorithm, suitable for cases where defects and background have significant brightness differencesComplex detection process, limited adaptability to illumination changesEL images
      Other featuresComprehensive consideration of multiple features, improving detection robustnessComplex feature selection and combination process, requires sufficient domain knowledge
      Clustering methodAutomatically discovers clustering structures in dataClustering parameters need targeted real-time adjustments
      Thresholding methodSimple implementation, suitable for simple scenesClustering parameters need targeted real-time adjustments
      Region growing methodObtains local structures, suitable for continuous defect areasSensitive to the selection of initial seeds, challenging for complex-shaped defects
      Transform domain analysisMatrix decomposition methodBetter differentiation of different types of defectsHigher requirements for image acquisitionDigital images
      Fourier transformGood performance for scratches, cracks, spots, and grid breakage, low complexity, good real-time performanceHigher requirements for image preprocessing, parameter setting needed
      Wavelet transform
      ICAGood performance for scratches, cracks, spots, and grid breakage, low complexity, good real-time performanceAlgorithm performance heavily relies on data preprocessing
    • Table 4. Advantages and disadvantages of specific methods of supervised learning

      View table
      View in Article

      Table 4. Advantages and disadvantages of specific methods of supervised learning

      NetworkMethodYearAdvantagesLimitations
      DetectionResNet50+Faster R-CNN+GA-RPN852019High detection accuracy, fast speedIncomplete coverage of battery types and defect types in the dataset
      Faster RPAN-CNN872020Good generality in detecting multiple defectsSuboptimal detection for multi-scale defects
      MF-RPN+Faster R-CNN842021Strong adaptability to defect shape and scale changesLong detection time, high algorithm complexity
      BAFPN-CNN882021Robust detection of solar cell crack scalesManual setting of feature balance factor in attention module
      DPiT+CW-MSA952021Strong feature extraction capability, good detection performanceRelatively large computational load
      BPGA-CNN912022Reduces the impact of complex background noise on small target defect detectionLimited types of defects in the dataset
      YOLO v5+CSP+ECA-Net922022High detection accuracy, good performanceHigh model complexity, slow detection speed
      Densenet121+YOLOv4902023Improved detection accuracy and speedModel performance needs further improvement
      YOLO v5+DCNv2+CA932023Good detection performance for small-sized defects, fewer model parametersUneven distribution of defect types in the dataset
      DCNN942023Quantifies the degree of defects, making the solar cell manufacturing process more intelligentLow resolution of images used due to poor image acquisition system quality
      SegmentationMAU-Net992020Effectively extracts salient features, good segmentation performanceRequires a well-annotated training dataset
      Pre-trained U-net1012020Directly obtains defect segmentation mapsDecreased pixel-level accuracy, longer processing time
      ERDCF-Net972022Effectively suppresses heterogenous texture background interference, grades solar cells based on damage levelUses only crack defect size as a reference for battery quality grading
      U2-Net+CSEB+DE-RSU1002023Finer segmentation of defect edgesLower resolution of images obtained using PL imaging technology
      HybridVGG16+U-Net++1032021Avoids overfitting, combines detection and segmentation tasksImprovement needed in defect localization effectiveness
      Faster R-CNN+EfficientNet+AE1042021Implements complete end-to-end networkModel robustness needs improvement
      LightweightShuffleNet V21052022Improved detection performance, practical utilityPoor detection performance for individual defect types
      YOLOv3-Tiny[107]2022Balances detection accuracy and speed, achieves real-time monitoringWeak model generalization ability
      Inception V3[106]2023High accuracy in defect recognition, fast classification speedLow pixel count in dataset images
      NAS+KD1082023Low hardware requirements, high detection accuracy, meets industrial application needClassifies only whether a battery has defects, cannot determine defect location and type
    • Table 4. Advantages and disadvantages ofw specific methods of supervised learning

      View table
      View in Article

      Table 4. Advantages and disadvantages ofw specific methods of supervised learning

      NetworkMethodYearAdvantagesLimitations
      ClassificationVGG-16+SVM832018Suitable for small datasetsPoor real-time detection
      Transfer Learning -VGG-19-352019Good accuracy on single crystal and uneven polycrystalline unitsOnly identifies defects, does not distinguish defect types
      Improved LeNet-5622019Higher accuracy and lower cross-entropy loss than classical LeNet-5 and SVM classifierLimited samples, focuses on one type of defect (cracks)
      SEF-CNN632019Features with better discriminability and robustnessLonger model training time
      ADI712020Detects defect types while judging the defect samplesRelatively low accuracy
      Modified AlexNet732020High accuracy for large-area defect detection, strong generalityOnly applicable to visible defects, not suitable for minor scratch detection
      HFCNN752020Low parameter usage, low memory occupation, eliminates data uncertaintyRelatively slow training speed
      L-CNN412021Simple network structure, fast training speedImbalanced class distribution in the dataset, model accuracy needs improvement
      Ensemble Learning CNN682021Improves detection accuracy while main-taining low costDetection accuracy and speed are not high enough
      PSO Pruner-CNN802022Reduces model complexitySlight decrease in model accuracy
    • Table 5. Advantages and disadvantages of unsupervised learning, weakly supervised and semi-supervised learning

      View table
      View in Article

      Table 5. Advantages and disadvantages of unsupervised learning, weakly supervised and semi-supervised learning

      AlgorithmYearAdvantagesLimitations
      Unsupervised learning201417Fast detection speed, some level of generalitySmall dataset, low image resolution
      2015112Avoids the impact of manually designed model parameters, strong interference resistance, high generalityLimited samples in the dataset
      2018113Exhibits a certain level of generalityLong training time, unable to handle high-resolution images
      2020114Model exhibits some robustness, strong feature expression capabilityWeak model generalization capability
      2020115Demonstrates good generalityCannot directly detect defects in the original image
      Weakly super-vised learning2019116Effectively segments based on a small number of annotationsLimited annotation information, rough segmentation results
      2020117Robustness and adaptability to complex surfaces without the need for extensive annotation data, has some generalityLimited annotation information, rough segmentation results
      2022118Model training requires only a small number of negative samples, strong generalization capabilityDoes not classify defect types
    • Table 6. Characteristics of various algorithms based on deep learning

      View table
      View in Article

      Table 6. Characteristics of various algorithms based on deep learning

      MethodNetworkOutput resultNetwork characteristicsEvaluation
      SupervisedClassificationDefect categoryOnly suitable for detecting the presence of single or multiple types of defects in solar cells, relatively low annotation costThe algorithm has strong learning ability for high-dimensional features, but computational complexity increases with the depth of the network
      DetectionDefect locationSuitable for classifying and locating multiple defects in a single image, higher annotation cost than classification networks
      SegmentationDefect shapeSuitable for classifying and locating multiple types of defects, pixel-level classification, provides specific defect shapes, higher annotation cost than classification and detection networks
      UnsupervisedGANDefect existenceGlobally trains the entire network, usability needs improvementSaves manual annotation costs, can detect unknown defects, but cannot classify defect types
      DBNDefect feature weightsCan generate data, network pre-trained layer by layer, strong feature extraction ability, but network parameters are limited by experience
      AEPredicted MaskStrong defect feature representation, higher robustness and accuracy than supervised learning, requires consistent input-output data dimensions
      Weakly supervised and semi-supervisedDefect category or locationLocalization performance is worse than supervised methods but better than unsupervised methodsSuitable for situations with imprecise data
    • Table 7. Solar cell surface defect detection public dataset

      View table
      View in Article

      Table 7. Solar cell surface defect detection public dataset

      DatasetSizeLink
      光伏异常检测392https://aistudio.baidu.com/aistudio/datasetdetail/183793
      PTID3 027http://vrai.dii.univpm.it/content/photovoltaic-thermal-images-dataset
      ELPV2 624https://github.com/zae-bayern/elpv-dataset
      ELDDS1400c51 400https://universe.roboflow.com/fma04-fayoum-edu-eg/eldds1400c5-dataset
      BELI593https://github.com/TheMakiran/BenchmarkELimages
      PVEL-AD36 543http://aihebut.com/col.jsp?id=118
      PVMD1 108https://github.com/CCNUZFW/PV-Multi-Defect
      ISM20 000https://github.com/RaptorMaps/InfraredSolarModules
      CS1 837https://datahub.duramat.org/dataset/crack-segmentation
    • Table 8. Evaluation metrics of classification and segmentation

      View table
      View in Article

      Table 8. Evaluation metrics of classification and segmentation

      MetricFormulaApplicability
      Accuracy↑Accuracy=TP+TNTP+TN+FP+FN

      Classification

      Segmentation

      Recall↑Recall=TPR=TPTP+FN
      Precision↑Precision=TPTP+FP
      F1-score↑F1-score=2*Precision×RecallPrecision+Recall
      FPR↓Falsealarm=FPR=FPTP+FP
      FNR↓Missrate=FNTP+FN=1-Recall
      IoU↑IoU=PGTPGT=TPTP+FP+FNSegmentation
      MIoU↑MIoU=1Cc=1CTPcTPc+FPc+FNc
      Dice coefficient↑Dice=2TP2TP+FP+FN
    • Table 9. Performance comparison and analysis of existing traditional machine vision algorithms

      View table
      View in Article

      Table 9. Performance comparison and analysis of existing traditional machine vision algorithms

      MethodDetection objectDetection approachSample SizePerformance
      Feature+RBF-SVM34Lack of angle, Broken gate, Collapse edge, crack, Leaking slurry, Casting pointClassification750 defect samplesRecognition rate: 90% and above
      KAZE+SVM35likelihood of defectsClassification

      2624 images, Size:

      300×300 pixels

      Accuracy 88.24%
      SVM39Parallel cracks, Series cracks, Bypass diode failure, Defect-free, Random defective solar cells.Classification

      100 defect-free images,

      100 images with defects,

      Pixel value: uint8

      Accuracy: 97.5%
      RBF+SVM40Appearance defects, Color defects, Cracks defects.Classification

      600 defect samples, Size:

      656×492 pixels

      Recognition rate: 98.57%, Computation time 0.107 s
      DBF+SVM41likelihood of defectsClassification

      2624 images, Size:

      300×300 pixels

      4 classes: 90.57%

      2 classes: 94.52%

      Haar feature +clustering45Micro-crack, break, finger-interruption.Pixel-level Segmentation

      31 defect-free images,

      19 images with defects

      Size: 550×550 pixels

      Processing time per image 0.1 s
      Low-rank representation51Surface defectPixel-level Segmentation

      40 images with defects,

      60 images without defects, Size: 226×226 pixels

      Precision: 0.867

      F1-score: 0.778

      Fourier transform54Micro-crack, break, finger-interruption.Pixel-level SegmentationImage size 550×550 pixels divided into 323 images of size 75×75, 308 defect-free samples, 15 samples with defectsRecognition rate: 1.00 Speed: 0.006 s per sub-image 0.29 s per original image
      Binary and Discrete Fourier Transform55Micro-crackClassification + Detection Window60 EL benchmark images, Size: 720×720 pixels

      Detection time for 60 cells

      High resolution: 1.62 s Low resolution: 2.52 s

      ICA58Internal defectPixel-level Segmentation28 defect-free images, 52 defective images, Size: 1 250×1 250 pixels0.04 seconds per image
    • Table 10. Comparison of the performance of existing representative algorithms on ELPV datasets

      View table
      View in Article

      Table 10. Comparison of the performance of existing representative algorithms on ELPV datasets

      MethodAccuracyPrecisionRecallF1-scoreDevice
      SVM-KAZE3582.44%82.52%Intel i7-3770 K CPU 32 GB RAM
      AC-VGG-193588.42%88.89%Two NVIDIA GeForce GTX 1080
      DFB-SVM4189.63%87.01%91.58%91.55%Intel Core i5 4200U 8 GB RAM
      L-CNN4182.58%80.71%96.59%85.80%Intel Xeon E-2224G 16 GB RAM
      VGG-11(Lightweight)6693.02%92.50%92.00%92.49%Intel Core i5,3.20 GHz CPU
      Inception-ResNet-v26993.00%1 NVIDIA TITAN GPU
      ADI7183.00%4 NVIDIA TITAN GPUs
      HFCNN97588.38%88.00%76.00%82.00%16 GB内存的NVIDIA TESLA P100 GPU
      RBF-SVM-SURF7772.74%Intel Xeon E5-2680 v4,2.4 GHz CPU
      Improved CNN7791.58%91.57%91.85%91.57%Nvidia GeForce GTX 1080 Ti
      SeF-HRNet7894.90%93.85%94.29%94.07%RTX 3090 GPU
      PSO Pruner-CNN8090.91%94.16%NVIDIA GeForce RTX 3080
      VGG-16+SVM8299.49%
      DPiT9591.70%93.20%79.60%85.90%8 GeForce GTX 1080Ti GPUs
      Improved VGG-1610395.20%95.40%94.40%NVIDIA TITAN GPU
      EfficientNet-B110484.00%16 GB GPU Tesla T4
      NAS+KD10891.74%87.13%86.28%86.70%Intel i9-10980XE 24.85 M
      Improved ResNet50-L111694.00%90.00%77.00%83.00%
    Tools

    Get Citation

    Copy Citation Text

    Yuqi LIU, Yiquan WU. Review of defect detection algorithms for solar cells based on machine vision[J]. Optics and Precision Engineering, 2024, 32(6): 868

    Download Citation

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

    Category:

    Received: Sep. 25, 2023

    Accepted: --

    Published Online: Apr. 19, 2024

    The Author Email: WU Yiquan (nuaaimage@163. com)

    DOI:10.37188/OPE.20243206.0868

    Topics