Research on a Rice Counting Algorithm Based on an Improved MCNN and a Density Map.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate rice grain counting is essential for determining thousand grain weight, a key indicator of crop quality and yield prediction.
• Existing methods face challenges with variations in rice seed shapes and adhesion, impacting counting accuracy.

Purpose of the Study:

• To develop an advanced method for precise rice grain counting using deep learning.
• To improve the accuracy of thousand grain weight estimation through enhanced seed detection.

Main Methods:

• Utilized an adaptive Gaussian kernel to generate accurate density maps from rice images.
• Employed a Multi-Column Convolutional Neural Network (MCNN) for feature extraction and fusion.
• Integrated an advanced prior step to estimate image density and mitigate adhesion effects.

Main Results:

• The proposed method demonstrated superior accuracy in rice grain counting compared to the original MCNN algorithm.
• The adaptive kernel and prior estimation effectively handled variations in rice shapes and adhesion.

Conclusions:

• The developed deep learning approach significantly enhances the accuracy of rice grain counting.
• This method provides a more reliable basis for thousand grain weight determination and agricultural yield prediction.