L2MXception: an improved Xception network for classification of peach diseases

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Peach production faces significant challenges from diseases, impacting yield and quality.
• Accurate and timely detection of peach diseases is crucial for effective management.
• Existing deep learning models struggle with limited and imbalanced image datasets for peach diseases.

Purpose of the Study:

• To develop an improved deep learning model for accurate peach disease detection.
• To address the challenges of data scarcity and imbalance in peach disease image datasets.
• To enhance the performance of deep learning models in identifying various peach diseases.

Main Methods:

• An improved Xception network, termed L2MXception, was proposed, incorporating L2-norm and mean regularization.
• A novel loss function, L2M Loss, was integrated into the deep learning model.
• The L2MXception model was trained and evaluated on a collected dataset of peach disease images.
• Performance was compared against seven mainstream deep learning models.

Main Results:

• L2MXception demonstrated superior performance compared to existing methods for peach disease prediction.
• The model achieved a validation accuracy of 93.85%, a substantial increase of 28.48% over the original Xception model.
• Experiments confirmed the effectiveness of L2M Loss in handling imbalanced datasets.

Conclusions:

• The L2MXception network shows significant potential for the early and accurate identification of peach diseases.
• This advancement can aid in improving peach crop management and reducing economic losses.
• The developed method offers a promising solution for applying deep learning to agricultural disease diagnostics.