A novel method for peanut variety identification and classification by Improved VGG16

Area of Science:

• Agricultural Science
• Computer Science
• Machine Learning

Background:

• Accurate crop variety identification is crucial for seed quality control, phenotype analysis, and scientific breeding.
• Traditional image processing methods for crop identification suffer from subjectivity and limited generalization.
• Peanut varieties exhibit distinct yield and quality traits, necessitating effective classification.

Purpose of the Study:

• To develop an improved deep convolutional neural network (VGG16) for enhanced crop variety identification.
• To apply the improved VGG16 model to classify 12 distinct varieties of peanuts.
• To evaluate the model's performance against established deep learning architectures.

Main Methods:

• Modified the VGG16 architecture by removing fully connected layers and adding new convolutional and pooling layers.
• Incorporated Batch Normalization (BN) layers and depth concatenation into the convolutional layers.
• Preprocessed 3365 peanut pod images across 12 varieties and fine-tuned the improved VGG16 model.

Main Results:

• The improved VGG16 model achieved an average accuracy of 96.7% for peanut pod variety identification.
• This accuracy surpassed the original VGG16 by 8.9% and other classical models by 1.6-12.3%.
• The model demonstrated robustness and generality, achieving 90.1% accuracy in classifying seven corn grain varieties.

Conclusions:

• The improved VGG16 model effectively identifies and classifies peanut pod varieties, validating the use of deep convolutional neural networks in this domain.
• The proposed model offers a significant advancement for crop variety identification and classification tasks.
• This research provides a feasible and effective approach for exploring other crop identification applications.