Detection of Sugarcane Leaf Diseases with A Hybrid Method of Different Deep Learning Models

Year 2025, Volume: 2 Issue: 1, 8 - 16, 29.07.2025

Mehmet Zahit Uzun

Abstract

Timely and accurate diagnosis of diseases encountered by sugar cane plants is crucial for agricultural activities. Traditional approaches have difficulties that require subjective, laborious, and expert judgment. In line with these needs, a deep learning-based model that automatically classifies diseases was developed in this study. The Sugarcane Leaf Disease dataset used in the research consists of 2521 images of 5 classes: red rot, mosaic, rust, yellow diseases, and healthy. The photos were multiplied using the data augmentation technique. In the study, a hybrid approach was preferred, and four different deep learning models were used. Transfer learning was performed using ImageNet weights in all models selected, including DenseNet, InceptionResNet, MobileNet, and Xception. For DenseNet and MobileNet, the image inputs are 224×224; for Xception and InceptionResNet, the input values are 299×299. In the models, the batch size is 32, global average pooling, and the activation function “relu” is used in fully connected layers. In the last layers of all models, first a dropout layer with a value of 0.5 was used, then a fully connected layer with 1000 neurons was used.1000 features from each model's last fully connected layer, a total of 4000 features, were classified with a Support Vector Machine (SVM). With Rbf kernel and C:100, gamma=0.001 hyperparameters, 92.61% accuracy, 92.66% precision, 92.54% sensitivity, and 92.57% F1 measurements were obtained. The RepeatedStratifiedKfold 5-fold cross-validation technique and GridCVSearch method, optimizing hyperparameters, were used in experimental studies. The results showed that the model could be a potential decision support system alternative to traditional methods.

Keywords

Classification , SVM , Deep learning , Sugarcane leaf disease , Feature extraction

Şeker Kamışı Yaprak Hastalıklarını Farklı Derin Öğrenme Modellerinin Hibrit Bir Yöntemi ile Tespiti

Abstract

Şeker kamışı bitkisinde karşılaşılan hastalıkların zamanında ve doğru teşhisi tarımsal faaliyetler açısından büyük önem taşımaktadır. Geleneksel yaklaşımların öznel, iş gücü ve uzman görüşü gerektiren zorlukları bulunmaktadır. Bu ihtiyaçlar doğrultusunda bu çalışmada hastalıkları otomatik olarak sınıflandıran derin öğrenme tabanlı bir model geliştirilmiştir. Araştırmada kullanılan Sugarcane Leaf Disease veri seti kırmızı çürüklük, mozaik, pas, sarı hastalıkları ve sağlıklı olmak üzere 5 sınıftan oluşan toplam 2521 adet görüntüden oluşmaktadır. Görüntüler veri artırma tekniği kullanılarak çoğaltılmıştır. Çalışmada hibrit yaklaşım tercih edilmiş ve dört farklı derin öğrenme modeli kullanılmıştır. DenseNet, InceptionResNetV2, MobileNet ve Xception olmak üzere seçilen tüm modellerde ImageNet ağırlıkları kullanılarak transfer öğrenmesi gerçekleştirilmiştir. DenseNet ve MobileNet için görüntü girişleri 224x224; Xception ve InceptionResNet için giriş değerleri 299x299'dur. Modellerde toplu boyut 32, global ortalama havuzlama ve tam bağlı katmanlarda aktivasyon fonksiyonu “relu” kullanılmıştır. Tüm modellerin son katmanlarında önce 0.5 değerine sahip bırakma katmanı ardından 1000 nöronlu bir tam bağlantılı katman kullanıldı. Her modelin son tam bağlantılı katmanından 1000 özellik, toplam 4000 özellik Destek Vektör Makinesi (DVM) ile sınıflandırıldı. Rbf çekirdeği ve C:100, gamma=0.001 hiper parametresi ile %92.61 doğruluk, %92.66 kesinlik, %92.54 duyarlılık ve %92.57 F1 ölçümleri elde edildi. Deneysel çalışmalarda RepeatedStratifiedKfold 5 katlı çapraz doğrulama tekniği ve hiper parametreleri optimize eden GridCVSearch yöntemi kullanıldı. Sonuçlar, modelin geleneksel yöntemlere alternatif olabilecek potansiyel bir karar destek sistemi olabileceğini gösterdi.

Keywords

Sınıflandırma , DVM , Derin Öğrenme , Şeker Kamışı Yaprağı , Özellik Çıkarma

References

• Başaran E (2022). A new brain tumor diagnostic model: Selection of textural feature extraction algorithms and convolution neural network features with optimization algorithms. Computers in Biology and Medicine, 148, 105857. https://doi.org/https://doi.org/10.1016/j.compbiomed.2022.105857.
• Chollet F (2017). Xception: Deep learning with depthwise separable convolutions. In Proceedings of the IEEE conference on computer vision and pattern recognition, 1251-1258.
• Daphal SD, Koli SM (2021). Transfer Learning approach to Sugarcane Foliar disease Classification with state-of-the-art Sugarcane Database. In 2021 International Conference on Computational Intelligence and Computing Applications (ICCICA), 1-4. https://doi.org/10.1109/ICCICA52458.2021.9697312.
• Daphal SD, Koli SM (2023). Enhancing sugarcane disease classification with ensemble deep learning: A comparative study with transfer learning techniques. Heliyon, 9(8). https://doi.org/10.1016/j.heliyon.2023.e18261.
• Daphal SD, Koli SM (2024a). Enhanced Classification of Sugarcane Diseases Through a Modified Learning Rate Policy in Deep Learning. Traitement du Signal, 41(1), 441-449. https://doi.org/10.18280/ts.410138.
• Daphal SD, Koli SM (2024b). Enhanced deep learning technique for sugarcane leaf disease classification and mobile application integration. Heliyon, 10(8). https://doi.org/10.1016/j.heliyon.2024.e29438.
• Daphal S, Koli S (2025, Mayıs 4). Sugarcane Leaf Disease . https://data.mendeley.com/datasets/9424skmnrk/1.
• Howard AG, Zhu M, Chen B, Kalenichenko D, Wang W, Weyand T, Andreetto M, Adam H (2017). Mobilenets: Efficient convolutional neural networks for mobile vision applications. arXivpreprint arXiv:1704.04861.
• Huang YK, Li WF, Zhang RY, Wang XY, Huang YK, Li WF, Zhang RY, Wang XY (2018). Diagnosis and control of sugarcane important diseases. In Color Illustration of Diagnosis and Control for Modern Sugarcane Diseases, Pests, and Weeds, 1-103.
• Hughes CG (1978). Diseases of Sugarcane — a Review. PANS, 24(2), 143-159. https://doi.org/10.1080/09670877809411604.
• Karadeniz AT, Başaran E, Celık Y (2023). Identification of walnut variety from the leaves using deep learning algorithms. Bitlis Eren Üniversitesi Fen Bilimleri Dergisi, 12(2), 531-543. https://doi.org/10.17798/bitlisfen.1263130.
• Karadeniz AT, Çelik Y, Başaran E (2023). Classification of walnut varieties obtained from walnut leaf images by the recommended residual block based CNN model. European Food Researchand Technology, 249(3), 727-738. https://doi.org/10.1007/s00217-022-04168-8.
• Li X, Li X, Zhang S, Zhang G, Zhang M, Shang H (2023). SLViT: Shuffle-convolution-based lightweight Vision transformer for effective diagnosis of sugarcane leaf diseases. Journal of King SaudUniversity -Computerand Information Sciences, 35(6), 101401. https://doi.org/https://doi.org/10.1016/j.jksuci.2022.09.013.
• Rajput K, Manwal M, Chauhan RK, Kukreja V, Mehta S (2024). Transforming Sugarcane Leaf Diseases Pathology with Convolutional Neural Networks and SVM. 2024 11th International Conference on Reliability, Infocom Technologies and Optimization (Trendsand Future Directions) (ICRITO), 1-6. https://doi.org/10.1109/ICRITO61523.2024.10522214.
• Rajput K, Manwal M, Kukreja V, Mehta S (2024). Enhancing Crop Health: CNN-SVM Fusion for Sugarcane Leaf Disease Analysis. 2024 3rd International Conference for Innovation in Technology (INOCON), 1-5. https://doi.org/10.1109/INOCON60754.2024.10511764.
• Ratnasari EK, Mentari M, Dewi RK, Ginardi RVH (2014). Sugarcane leaf disease detection and severity estimation based on segmented spots image. In Proceedings of International Conference on Information, Communication Technology and System (ICTS) 2014, 93-98. https://doi.org/10.1109/ICTS.2014.7010564.
• Singh R, Sharma N, Pal A (2024). Optimized Deep Learning Approaches for Sugarcane Leaf Disease Classification Using CNN Model. 2024 5th International Conference on Data Intelligence and Cognitive Informatics (ICDICI), 598-603. https://doi.org/10.1109/ICDICI62993.2024.10810874.
• Sun C, Zhou X, Zhang M, Qin A (2023). SE-VisionTransformer: Hybrid network for diagnosing sugarcane leaf diseases based on attention mechanism. Sensors, 23(20), 8529.
• Szegedy C, Ioffe S, Vanhoucke V, Alemi A (2017). Inception-v4, inception-resnet and the impact of residual connections on learning. In Proceedings of the AAAI conference on artificial intelligence, 31(1).
• Tonkal Ö, Polat H, Başaran E, Cömert Z, Kocaoğlu R (2021). Machine learning approach equipped with neighbourhood component analysis for DDoS attack detection in software-defined networking. Electronics, 10(11). https://doi.org/10.3390/electronics10111227.