Position-sensitive Neural Networks For Insect Detection And Grain Quality Prediction In Storage

Abstract

Insect Infestations In Grain Can Lead To Significant Losses In Both Quantity And Quality, Impacting Crop Value. These Insects Not Only Consume Grain But Also Contaminate It With Their Metabolic By-products And Body Parts, Contributing To The Growth Of Microflora And The Creation Of Hotspots Due To The Heat And Moisture Generated By Their Activity. Severely Infested Grains Are Unsuitable For Seed Purposes And Their Products Are Unfit For Human Consumption. As Such, Effective Monitoring And Detection Of Stored-product Insects Is Crucial. Recent Advancements In Hardware Computing Have Led To Notable Progress In Deep Learning-based Computer Vision Techniques For Object Detection, Including The Detection Of Insects On Grain Surfaces. Many Grain Depots Now Utilize High-definition Cameras And Insect-monitoring Systems That Capture Images Or Videos, Offering A Practical Opportunity For Deep Learning Models To Assist In Detecting Insect Infestations. This Project Proposes An Enhanced Neural Network Architecture Based On Incremental Learning Networks To Detect And Classify Eight Common Stored Grain Insect Species And Predict Grain Severity. The Proposed Architecture Incorporates A Neural Network For Feature Extraction, A Region Proposal Network, And A Position-sensitive Score Map For Improved Target Detection. By Integrating A Position-sensitive Score Map In Place Of Some Fully Connected Layers, The Network Becomes More Adaptable To Complex Backgrounds, Enabling Faster And More Accurate Insect Detection. This Innovative Architecture Also Introduces Position-sensitive ROI Pooling To Further Improve Performance. Experimental Results Demonstrate That The Proposed Model Significantly Outperforms Existing Models, Achieving Higher Precision-recall Rates For Insect Detection In Grain Images. The Proposed Solution Offers An Effective And Efficient Method For Monitoring Insect Infestations In Stored Grain, Ensuring Better Crop Quality And Minimizing Losses.

Keywords