Evaluation of segmentation methods for RGB colour image-based detection of Fusarium infection in corn grains using support vector machine (SVM) and pre-trained convolution neural network (CNN)

Canadian Biosystems Engineering

article

T.S. Rathna Priya¹  Annamalai Manickavasagan¹ 

[1] School of Engineering, University of Guelph

关键词: corn;    imaging;    segmentation;    classification;    algorithm;    accuracy;    Fusarium;   

DOI  :  10.7451/CBE.2022.64.7.9

学科分类：农业科学（综合）

来源: Canadian Society of Agricultural Engineering

【 摘 要 】

This study evaluated six segmentation methods (clustering,flood-fill, graph-cut, colour-thresholding, watershed, andOtsu’s-thresholding) for segmentation accuracy andclassification accuracy in discriminating Fusarium infectedcorn grains using RGB colour images. The segmentationaccuracy was calculated using Jaccard similarity index andDice coefficient in comparison with the gold standard(manual segmentation method). Flood-fill and graph-cutmethods showed the highest segmentation accuracy of 77%and 87% for Jaccard and Dice evaluation metrics,respectively. Pre-trained convolution neural network(CNN) and support vector machine (SVM) were used toevaluate the effect of segmentation methods onclassification accuracy using segmented images andextracted features from the segmented images, respectively.The SVM based two-class model to discriminate healthyand Fusarium infected corn grains yielded the classificationaccuracy of 84%, 79%, 78%, 74%, 69% and 65% for graphcut, watershed, clustering, flood-fill, colour-thresholding,and Otsu’s-thresholding, respectively. In pretrained CNNmodel, the classification accuracies were 93%, 88%, 87%,84%, 61% and 59% for flood-fill, graph-cut, colourthresholding, clustering, watershed, and Otsu’sthresholding, respectively. Jaccard and Dice evaluationmetrics showed the highest correlation with the pretrainedCNN classification accuracies with R2 values of 0.9693 and0.9727, respectively. The correlation with SVMclassification accuracies were R2–0.505 for Jaccard and R2–0.5151 for Dice evaluation metrics.

【 授权许可】

CC BY