Identification method for multiple similar crop pests: A visual language model combined with coarse-finegrain...

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Crop pests and diseases pose significant agricultural challenges.
• Identifying pests is difficult due to similar characteristics and complex field environments.
• Conventional single-modal approaches struggle with dynamic surroundings.

Purpose of the Study:

• To develop an advanced cross-modal model for accurate crop pest identification.
• To address limitations of existing methods in complex agricultural settings.
• To enhance the detection of crop pests and diseases.

Main Methods:

• A two-phase cross-modal model was developed for pest identification.
• Phase one used coarse- and fine-grained cross-modal matching on image-text pairs.
• Phase two employed a 'point-to-area' module for key feature super-localization.

Main Results:

• The model achieved improved classification accuracy compared to previous cross-modal models (11.68%, 4.26%, 5.90% increases).
• Validation on IP102 and Li and Xie datasets confirmed effectiveness.
• Experimental results demonstrate suitability for complex environments.

Conclusions:

• The proposed model effectively identifies crop pests in challenging real-world conditions.
• It offers a significant advancement over existing cross-modal identification techniques.
• The model is suitable for practical crop pest and disease monitoring and detection.