Convolutional neural network in rice disease recognition: accuracy, speed and lightweight

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Rice diseases significantly impact crop yield and quality, necessitating accurate and efficient identification methods.
• Traditional manual disease identification relies heavily on expert experience, limiting scalability and consistency.
• Convolutional neural networks (CNNs) offer advanced image analysis capabilities for feature extraction in complex biological data.

Purpose of the Study:

• To review the application of CNN technology in automated rice disease recognition.
• To highlight advancements in recognition accuracy, processing speed, and mobile deployment of CNN models for rice diseases.
• To discuss strategies for optimizing CNNs, including lightweighting for real-time applications and dataset/model improvements.

Main Methods:

• Review of existing literature on CNN applications in rice disease identification.
• Analysis of CNN architectures, including ensemble and transfer learning techniques.
• Exploration of model lightweighting and dataset enhancement strategies for improved performance.

Main Results:

• CNNs have demonstrated significant improvements in rice disease recognition accuracy and speed compared to manual methods.
• The development of lightweight CNN models facilitates real-time disease detection and mobile deployment in agricultural settings.
• Enhancements in datasets and model structures further boost the performance of automated rice disease recognition systems.

Conclusions:

• CNN technology provides a powerful and scalable solution for automated rice disease identification, reducing labor costs and increasing accuracy.
• Ongoing research focuses on optimizing CNN models for real-time, mobile-based applications to support precision agriculture.
• Further improvements in datasets and model architectures will continue to enhance the effectiveness of CNNs in managing rice diseases.