Denoising and Segmentation in Pest Images Using Advanced Neural Networks

Abstract

In agricultural research and pest management, high-quality images play a crucial role in accurate analysis and decision-making. However, these images are often plagued by noise and clutter, which can hinder the effectiveness of automated analysis algorithms. This study proposes a two-phase approach for enhancing pest images by addressing noise and improving segmentation accuracy. The first phase employs a Mono Noise Elimination Neural Network (MNENN) designed specifically for denoising agricultural images. MNENN effectively removes noise and artifacts, enhancing the clarity and quality of pest images. This denoising step is crucial for ensuring accurate analysis in subsequent stages. In the second stage, an Enhanced Mask-aware Region-based Convolutional Neural Network (RCNN) is utilized for precise segmentation of pests from the denoised images. The enhanced RCNN incorporates mask-aware techniques to improve segmentation accuracy, particularly in complex backgrounds or overlapping pest scenarios. By utilizing advanced neural networks and segmentation strategies, this research aims to provide clean and accurately segmented pest images, facilitating further analysis and decision-making in pest management applications.