DeeppestNet: An end-to-end framework for high-performance crop pest recognition.

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Accurate crop-pest identification is crucial for sustainable agriculture but is hindered by noisy backgrounds and limited data.
• Existing methods struggle with feature extraction and algorithmic complexity in pest recognition.

Purpose of the Study:

• To propose DeeppestNet, a novel graph pyramid attention-based Bidirectional Long Short-Term Memory (GPA-BiLSTM) model for accurate and efficient crop-pest identification.
• To address challenges of noisy backgrounds, limited data, and algorithmic complexity in pest recognition.

Main Methods:

• Image quality enhancement using Contrast Adaptive Limited Histogram Equalization (CLAHE).
• Feature extraction using an adaptive pyramid attention module with a cross stage partial (AP-CSP) backbone and a multi-level pyramid structure.
• Classification using Graph-based BiLSTM (G-BiLSTM) optimized with Grey Wolf-Salp Swarm Optimization (GW-SSO).

Main Results:

• DeeppestNet achieved 4.6% higher accuracy than EfficientNet on the IP-102 dataset.
• The proposed method demonstrated superior performance compared to state-of-the-art Deep Learning algorithms.
• The model showed accuracy, efficiency, and computational efficiency.

Conclusions:

• DeeppestNet offers an accurate and efficient solution for crop-pest identification, overcoming limitations of previous methods.
• The integration of CLAHE, AP-CSP, G-BiLSTM, and GW-SSO effectively enhances pest recognition capabilities.
• The proposed model contributes to advancing sustainable agriculture through improved pest management technologies.