A Novel Transfer Learning Approach for Detection of Pomegranates Growth Stages

In this paper, we introduce a novel method, CRnet, designed to detect and classify the various growth stages of pomegranates using an integrated transfer learning approach. CRnet combines the strengths of Convolutional Neural Networks (CNN) and Random Forest (RF) classifiers to achieve accurate and efficient feature extraction and classification. Initially, spatial features are extracted from images of pomegranate growth stages using a pre-trained ResNet-18 CNN. These features are subsequently passed into a Random Forest model to classify the growth stages. The proposed technique demonstrates improved performance by leveraging CNN's ability to extract deep, high-dimensional spatial features and RF's robustness in classification. The dataset used comprises images of pomegranate plants, processed with noise removal, contrast adjustment, and segmentation for feature extraction. A comparative analysis is conducted to evaluate the model's accuracy, precision, recall, and computational complexity, achieving promising results. The method effectively enhances the classification of growth stages, contributing to automated agricultural monitoring and management. Experimental results indicate that CRNet can accurately predict growth stages with significant speed and reliability, positioning it as a valuable tool for pomegranate growers and researchers in precision agriculture.

Keywords: Pomegranates Dataset, Image Processing Techniques, Deep Learning Techniques, Machine Learning Techniques, Classification, Accuracy.