[Paper Review] An Effective Scheme for Maize Disease Recognition based on Deep Networks

Saeedeh Osouli, Behrouz Bolourian Haghighi+1 more|arXiv (Cornell University)|May 9, 2022

Smart Agriculture and AI4 citations

TL;DR

This paper proposes Mob-INC, a novel deep learning model combining fine-tuned MobileNetV2 and Inception modules via transfer learning for accurate maize disease recognition. By leveraging feature extraction from both architectures and addressing class imbalance with data augmentation, the model achieves 97% accuracy on a multi-class maize disease dataset, outperforming state-of-the-art methods.

ABSTRACT

In the last decades, the area under cultivation of maize products has increased because of its essential role in the food cycle for humans, livestock, and poultry. Moreover, the diseases of plants impact food safety and can significantly reduce both the quality and quantity of agricultural products. There are many challenges to accurate and timely diagnosis of the disease. This research presents a novel scheme based on a deep neural network to overcome the mentioned challenges. Due to the limited number of data, the transfer learning technique is employed with the help of two well-known architectures. In this way, a new effective model is adopted by a combination of pre-trained MobileNetV2 and Inception Networks due to their effective performance on object detection problems. The convolution layers of MoblieNetV2 and Inception modules are parallelly arranged as earlier layers to extract crucial features. In addition, the imbalance problem of classes has been solved by an augmentation strategy. The proposed scheme has a superior performance compared to other state-of-the-art models published in recent years. The accuracy of the model reaches 97%, approximately. In summary, experimental results prove the method's validity and significant performance in diagnosing disease in plant leaves.

Motivation & Objective

• To address the challenge of accurate and timely diagnosis of maize leaf diseases, which threaten food security and reduce crop yield.
• To overcome limitations in existing methods, such as low accuracy, high cost of expert diagnosis, and inefficiency in large-scale farming.
• To develop an automated, cost-effective, and precise system for classifying maize diseases using deep learning.
• To improve performance on small, imbalanced datasets through transfer learning and data augmentation strategies.

Proposed method

• The proposed model, Mob-INC, integrates pre-trained MobileNetV2 and Inception modules as parallel early convolutional layers to extract hierarchical features from maize leaf images.
• Transfer learning is applied by fine-tuning the entire MobileNetV2 network on the maize disease dataset after initial pre-training on ImageNet.
• An Inception module is appended after the shared feature extraction branches to enhance feature representation and capture multi-scale patterns.
• Class imbalance is mitigated using an augmentation strategy that increases minority class samples to improve model generalization.
• A new fully connected layer is added on top of the fused features for final disease classification.
• The model is trained for 30 epochs using a 70/10/20 train/validation/test split, with early stopping based on validation loss.

Experimental results

Research questions

• RQ1 Can a hybrid CNN architecture combining MobileNetV2 and Inception modules improve classification accuracy for maize disease detection compared to standalone models?
• RQ2 How effective is transfer learning with data augmentation in handling small and imbalanced maize disease datasets?
• RQ3 To what extent does the proposed Mob-INC model outperform existing state-of-the-art models in terms of accuracy and robustness?
• RQ4 Can the model achieve high confidence predictions across all disease classes, including those with similar visual features?

Key findings

• The Mob-INC model achieved a test accuracy of 97%, significantly outperforming other state-of-the-art models reported in recent literature.
• The model reached a validation accuracy of 96.10% and a validation loss of 0.14 after 30 training epochs.
• The confusion matrix revealed high classification confidence, with only minor misclassifications observed for three primary disease classes.
• Precision, F1-Score, Sensitivity, and Specificity metrics across all classes confirmed strong and balanced performance.
• Probability outputs for each disease class showed high confidence with minimal overlap between classes, indicating reliable decision boundaries.
• The ablation study confirmed that the combination of MobileNetV2 and Inception modules contributed to superior feature learning and overall model performance.