A methodical analysis of deep learning techniques for detecting Indian lentils

Abstract

Automatic identification of Indian lentils requires the selection of a suitable classification model among various CNN models. The CNN models are the most effective for image classification. However, more than one CNN model occasionally performs closer to each other. Choosing the best classification model requires considerable effort in this situation. Automated identification of lentils would allow for faster selection processes and reduced time compared to manual assessment. In this study, 18 CNN models were tested to identify the lentils. The performance of the CNN models was investigated and a two-phase statistical analysis was conducted to select the best CNN model for identifying Indian lentils. The 18 CNN models are Alexnet, Darknet19, Darknet53, Densenet, EfficientNetB0, Google net, InceptionResnetV2, InceptionV3, MobilenetV2, NasnetLarge, NasnetMobile, Resnet18, Resnet50, Resnet101, Squeezenet, Vgg16, Vgg19, Xception. The two-phase statistical analysis was Duncan's multiple range test, and the Wilcoxon signed-rank test was performed. Again, nine measures, such as accuracy, sensitivity, specificity, precision, FPR, F1 Score, MCC, Kappa, and computational time, were considered for statistical analysis. After the execution of 18 CNN models and two-phase statistical analysis, it was revealed that EfficientNetB0 is superior among the CNN models for the identification of lentils.

Keywords

• Lentils identification
• Deep learning
• Statistical analysis
• CNN
• Grain