Classification of Walnut Leaf Images Using a Hybrid CNN-Based Deep Learning Approach

Authors:
Publisher website
Google Scholar
Abstract
Walnut is a widely cultivated crop with various types and qualities, offering significant health benefits. However, its long production cycle and high cultivation costs necessitate the selection of appropriate varieties for specific ecological conditions. Due to morphological and color similarities, differentiating walnut varieties remains challenging, even for experts. Existing studies on walnut classification are limited and mostly confined to laboratory-based experiments. In this study, a novel hybrid computer-based approach is proposed for the automatic classification of walnut varieties using leaf images. A dataset consisting of 1,751 images from 18 different walnut varieties was collected from the Atatürk Horticultural Central Research Institute in Yalova, Turkey. The proposed model integrates deep features extracted from lightweight Convolutional Neural Network (CNN) architectures, namely SqueezeNet and MobileNetV2, with textural features obtained through the Gray-Level Co-occurrence Matrix (GLCM). The most significant features were selected using the Chi-square test, and classification was performed with Support Vector Machines (SVM). Experimental results demonstrate that the proposed hybrid model achieved an accuracy of 84.75% in classifying walnut varieties. These findings indicate that the proposed method can provide a reliable, fast, and cost-effective solution for walnut variety identification, with potential benefits for agricultural standardization and precision farming practices.

This publication has 23 references indexed in Scilit: