This study comparatively investigates the performance of deep learning and hybrid approaches for the detection and classification of potato leaf diseases (early blight, late blight, and healthy). In the first stage, direct image classification was performed using pre-trained deep learning models DenseNet201, ResNet50V2, VGG16, and Xception. Of these models, the VGG16 model achieved the highest accuracy. In the second stage, the same deep learning models were used as feature extractors, and the resulting features were classified using traditional machine learning algorithms, SVM, KNN, RF, and XGB. These hybrid approaches provided a significant increase in classification performance. The findings revealed that DenseNet201's combination of SVM and XGB exhibited superior performance with an overall accuracy rate of 99.31%. These results demonstrate that the powerful feature extraction capabilities of deep learning architectures, combined with the effective classification power of traditional machine learning algorithms, provide higher accuracy and reliability compared to the direct deep learning approach. The study highlights the potential of hybrid approaches, particularly for applications such as agricultural image processing and plant disease detection.
Şükrü Aykat (Wed,) studied this question.