Figures (7)  Tables (2)
    • Figure 1. 

      The characteristics that are evaluated in different seed classes during seed certification process.

    • Figure 2. 

      A schematic illustration of an artificial neuron.

    • Figure 3. 

      ANNs algorithms from old to new versions.

    • Figure 4. 

      Schematic views of SVMs, DA, NB and RF algorithms.

    • Figure 5. 

      Schematic views of ML applications in seed recognition, classification and grading.

    • Figure 6. 

      Schematic illustration of weed seeds identification in seed lots by ANN.

    • Figure 7. 

      application of ML models in seeds' viability.

    • ML algorithmsSOMMLPDASVMsRFNBCNNs
      Accuracy++++++++++++++++++++++++
      Flexibility++++++++++++++++++++
      AdvantagesFastNon-linear classificationQuick, inexpensiveAnalyzing complex networks, diminishing the generalization error, using large number of hidden unitsTolerant of
      highly correlated predictors
      Simplicity,
      fast training, decreases the number of parameters
      Analysis of massive
      amounts of
      unsupervised data, better classification
      and prediction
      DisadvantagesUnsupervisedTime-consuming, few hidden neuronsUnsupervisedAlgorithmic complexity, development of ideal classifiers for multi-class problems and unbalanced data setsEvaluation
      of pairwise interactions is difficult, future predictions require the original data
      Oversensitive
      to redundant
      or irrelevant attributes, classification
      bias
      It needs further developments
      for big data
      analysis

      Table 1. 

      Comparisons among ML algorithms.

    • Plant speciesType of machine learningClassifierAccuracyFeaturesPurposeRef.
      CornDigital imageMLP98.83%Texture - spectrum hybridSeed varietal purity[10]
      WheatDigital imageANNs85.72%MorphologySeed varietal purity[7]
      WheatDigital imageICA-ANN hybrid96.25%Color, morphology,
      and texture
      Seed varietal purity[72]
      WheatDigital bulk imageLDA
      98.15%
      TextureSeed varietal purity[37]
      WheatDigital bulk imageANNs97.62%TextureSeed varietal purity[73]
      Forage grass (Urochloabrizantha)FT-NIR spectroscopy & X-ray imagingRF85%Spectrum-composition hybridSeed germination & vigor[74]
      CornFT-NIR spectroscopyPLS-DA100%Chemical compositionSeed germination & vigor[75]
      PepperFT-NIR & Raman spectroscopyPLS-DA99%Chemical compositionSeed germination & vigor[76]
      57 weed speciesDigital imageNB & ANNs99.5%Color, morphology,
      and texture
      Weed identification[59]
      WheatVideo processingANN - PSO hybrid97.77%Shape, texture & colorPhysical purity & weed identification[77]
      RiceDigital imageMLP99.46%Morphology, texture & colorSeed varieties classification[9]
      RiceDigital imageANNsMorphologySeed grading[78]
      RiceDigital imageDFA96%MorphologyPhysical purity[79]
      SoybeanAerial imageryCNNs65%Object detectionWeed identification[80]
      SoybeanDigital imageCNNs97%Color, texture and shapeSeed deficiency[66]
      SoybeanDigital imageCNNs86.2%Color, texture and shapeSeed counting[70]
      CornDigital imageMLP94.5%ColorPhysical purity[81]
      SoybeanFlatbed scannerSVMs & RF78%ColorSeed grading[47]
      BeanDigital imageRF95.5%Color, texture and shapeSeed varieties classification[16]
      CornHyperspectral imageSVMs98.2%Spectrum-texture – morphology hybridSeed varieties classification[48]
      CornDigital imageSVMs95.6%Color and textureSeed varieties classification[49]
      CornDigital imageGA–SVM hybrid94.4%Color, texture and shapeSeed varieties identification[50]
      CornDigital imageCNNs95%Color, texture and shapeHaploid and dioploid discrimination[82]
      CornDigital imageCNNs95%Color, texture and shapeSeed varieties identification[83]
      BarleyDigital imageDA & K-NN99%Color, morphology & textureSeed varieties classification[8]
      BarleyDigital imageCNNs93%Color, morphology & textureSeed varieties classification[84]
      MLP: Multilayer Perceptron, ICA: Imperialist Competitive Algorithm, ANNs: Artificial Neural Networks, LDA: linear discriminate analysis, FT-NIR: Fourier transform near-infrared, PLS-DA: partial least squares discriminant analysis, NB: naïve Bayes, PSO: partial swarm optimization, DFA: stepwise discriminant function analysis, CNNs: Convolutional neural networks, SVM: support vector machine, GA: genetic algorithm, DA: discriminant analysis, K-NN: K-nearest neighbors, SVMs: support vector machines.

      Table 2. 

      Examples of applied machine learning models in modern seed recognition and classification studies.