Mobile Device-Based Detection System of Diseases and Pests in Rose Plants Using Deep Convolutional Neural Networks and Quantization

Burhan Duman

Makale detayları

Mobile Device-Based Detection System of Diseases and Pests in Rose Plants Using Deep Convolutional Neural Networks and Quantization

Yazarlar:

Burhan Duman

Yayıncı:

Ankara University

Ana Başlıklar:

Artificial Intelligence (Other) | Precision Agriculture Technologies

Dergi:

Journal of Agricultural Sciences Vol.31 Issue 2, pp.302-318

Belge Tipi:

Makale

Yayın Tarihi:

3/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001342995

pISSN:

1300-7580

Anahtar kelimeler:

Özet:

 In agriculture, the rapid and accurate identification of plant diseases and pests is crucial for maintaining the quality and yield of agricultural products. This study focuses on detecting diseases and pests affecting Rosa damascena Mill. plants through an ensemble learning approach and deploying the model in an Android mobile application-a rarity in similar research. A new dataset was created using images from the natural habitat and season of Rosa damascena Mill., covering seven different diseases and pests. For this approach, pre-training was performed with mixed- Convolutional Neural Network CNN models DenseNet169, ResNet152, MobileNetV2, VGG19, and NasNet. DenseNet169 and MobileNetV2, which are the models with the highest classification success obtained from mixed-CNN models, were combined in the new model by fine- tuning with the ensemble learning method. In the performance tests of the model, an accuracy of 95.17 was obtained. In addition, this study introduces an Android mobile application integrating these models, a distinctive feature compared to other similar studies. The best performances of these models, DenseNet169 and MobileNetV2 in both flat buffered and quantized forms, were performed separately on a computer, a physical mobile device, and an Android emulator. MobileNetV2 outperformed DenseNet169 2271 ms by having the lowest average inference time 301 ms on mobile devices. These results demonstrate the effectiveness of using a mobile device to detect rose plant diseases and pests efficiently in natural environments.

 In agriculture, the rapid and accurate identification of plant diseases and pests is crucial for maintaining the quality and yield of agricultural products. This study focuses on detecting diseases and pests affecting Rosa damascena Mill. plants through an ensemble learning approach and deploying the model in an Android mobile application-a rarity in similar research. A new dataset was created using images from the natural habitat and season of Rosa damascena Mill., covering seven different diseases and pests. For this approach, pre-training was performed with mixed- Convolutional Neural Network CNN models DenseNet169, ResNet152, MobileNetV2, VGG19, and NasNet. DenseNet169 and MobileNetV2, which are the models with the highest classification success obtained from mixed-CNN models, were combined in the new model by fine- tuning with the ensemble learning method. In the performance tests of the model, an accuracy of 95.17 was obtained. In addition, this study introduces an Android mobile application integrating these models, a distinctive feature compared to other similar studies. The best performances of these models, DenseNet169 and MobileNetV2 in both flat buffered and quantized forms, were performed separately on a computer, a physical mobile device, and an Android emulator. MobileNetV2 outperformed DenseNet169 2271 ms by having the lowest average inference time 301 ms on mobile devices. These results demonstrate the effectiveness of using a mobile device to detect rose plant diseases and pests efficiently in natural environments.

 In agriculture, the rapid and accurate identification of plant diseases and pests is crucial for maintaining the quality and yield of agricultural products. This study focuses on detecting diseases and pests affecting Rosa damascena Mill. plants through an ensemble learning approach and deploying the model in an Android mobile application-a rarity in similar research. A new dataset was created using images from the natural habitat and season of Rosa damascena Mill., covering seven different diseases and pests. For this approach, pre-training was performed with mixed- Convolutional Neural Network CNN models DenseNet169, ResNet152, MobileNetV2, VGG19, and NasNet. DenseNet169 and MobileNetV2, which are the models with the highest classification success obtained from mixed-CNN models, were combined in the new model by fine- tuning with the ensemble learning method. In the performance tests of the model, an accuracy of 95.17 was obtained. In addition, this study introduces an Android mobile application integrating these models, a distinctive feature compared to other similar studies. The best performances of these models, DenseNet169 and MobileNetV2 in both flat buffered and quantized forms, were performed separately on a computer, a physical mobile device, and an Android emulator. MobileNetV2 outperformed DenseNet169 2271 ms by having the lowest average inference time 301 ms on mobile devices. These results demonstrate the effectiveness of using a mobile device to detect rose plant diseases and pests efficiently in natural environments.

 In agriculture, the rapid and accurate identification of plant diseases and pests is crucial for maintaining the quality and yield of agricultural products. This study focuses on detecting diseases and pests affecting Rosa damascena Mill. plants through an ensemble learning approach and deploying the model in an Android mobile application-a rarity in similar research. A new dataset was created using images from the natural habitat and season of Rosa damascena Mill., covering seven different diseases and pests. For this approach, pre-training was performed with mixed- Convolutional Neural Network CNN models DenseNet169, ResNet152, MobileNetV2, VGG19, and NasNet. DenseNet169 and MobileNetV2, which are the models with the highest classification success obtained from mixed-CNN models, were combined in the new model by fine- tuning with the ensemble learning method. In the performance tests of the model, an accuracy of 95.17 was obtained. In addition, this study introduces an Android mobile application integrating these models, a distinctive feature compared to other similar studies. The best performances of these models, DenseNet169 and MobileNetV2 in both flat buffered and quantized forms, were performed separately on a computer, a physical mobile device, and an Android emulator. MobileNetV2 outperformed DenseNet169 2271 ms by having the lowest average inference time 301 ms on mobile devices. These results demonstrate the effectiveness of using a mobile device to detect rose plant diseases and pests efficiently in natural environments.

Makale detayları

Mobile Device-Based Detection System of Diseases and Pests in Rose Plants Using Deep Convolutional Neural Networks and Quantization

Yazarlar:

Burhan Duman

Yayıncı:

Ankara University

Ana Başlıklar:

Artificial Intelligence (Other) | Precision Agriculture Technologies

Dergi:

Journal of Agricultural Sciences Vol.31 Issue 2, pp.302-318

Belge Tipi:

Makale

Yayın Tarihi:

3/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001342995

pISSN:

1300-7580

Mobile aplication

Image classification

Deep learning

Plant disease and pests

Rosa damascena Mill.

Sayı

Makale detayları

Mobile Device-Based Detection System of Diseases and Pests in Rose Plants Using Deep Convolutional Neural Networks and Quantization

Yazarlar: Burhan Duman

Yayıncı: Ankara University

Ana Başlıklar: Artificial Intelligence (Other) | Precision Agriculture Technologies

Dergi: Journal of Agricultural Sciences Vol.31 Issue 2, pp.302-318

Belge Tipi: Makale

Yayın Tarihi: 3/2025

Belge Dili: İngilizce

Document ID (Org No.): 2001342995

pISSN: 1300-7580

Mobile aplication

Image classification

Deep learning

Plant disease and pests

Rosa damascena Mill.

Sayı

Yazarlar:

Burhan Duman

Yayıncı:

Ankara University

Ana Başlıklar:

Artificial Intelligence (Other) | Precision Agriculture Technologies

Dergi:

Journal of Agricultural Sciences Vol.31 Issue 2, pp.302-318

Belge Tipi:

Makale

Yayın Tarihi:

3/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001342995

pISSN:

1300-7580