Classification of Plant Leaves Using New Compact Convolutional Neural Network Models.

Area of Science:

• Agricultural technology
• Computer vision
• Machine learning

Background:

• Automated plant detection and classification are crucial for precision agriculture and crop safety.
• Existing methods may lack efficiency in terms of computational resources and training time.

Purpose of the Study:

• To develop and evaluate compact convolutional neural networks (CNNs) for plant species detection and classification.
• To compare the performance of proposed CNN models against AlexNet using transfer learning.
• To assess the impact of data augmentation on classification accuracy.

Main Methods:

• Development of novel compact CNN architectures (N1, N2, N3).
• Training models on the PlantVillage dataset (9 species) and Flavia dataset (32 classes) with data augmentation.
• Utilizing transfer learning with AlexNet as a benchmark.
• Evaluating models based on classification accuracy, training time, and model size.

Main Results:

• Proposed models achieved high classification accuracies (N1: 99.45%, N2: 99.65%, N3: 99.55%) on the PlantVillage dataset.
• Models demonstrated significant reductions in training time (up to 34.58%) and model size (up to 92.67%) compared to AlexNet.
• Data augmentation led to substantial improvements in classification accuracy.
• Models successfully classified plant leaves and identified diseases in tomato plants.

Conclusions:

• Compact CNN models offer an efficient alternative to AlexNet for plant detection and classification tasks.
• The developed models are suitable for resource-constrained environments in precision agriculture.
• These findings contribute to the advancement of automated crop monitoring and disease identification systems.