Variety Identification of Corn Seeds Based on Hyperspectral Imaging and Convolutional Neural Network

Area of Science:

• Agricultural Science
• Spectroscopy
• Machine Learning

Background:

• Corn seed variety identification is challenging due to visual similarities.
• Accurate classification is crucial for precision breeding and smart agriculture.
• Hyperspectral imaging offers non-destructive analysis for seed classification.

Purpose of the Study:

• To develop a rapid and non-destructive method for classifying 30 corn varieties from Northwest China.
• To minimize information loss and manual intervention in seed variety identification.
• To evaluate the performance of a novel deep learning model against traditional methods.

Main Methods:

• Analysis of hyperspectral images (870-1709 nm) from the embryonic region of corn seeds.
• Preprocessing using first-order derivatives to reduce noise and irrelevant information.
• Classification using KNN, ELM, RF, 1DCNN, and a CLA-CA (1DCNN-LSTM-ATTENTION-ECA) model.

Main Results:

• The CLA-CA model achieved the highest classification accuracy of 95.38%.
• The proposed model significantly outperformed traditional machine learning and 1DCNN models.
• Full-band spectral data and derivative preprocessing minimized information loss.

Conclusions:

• The CLA-CA model provides a highly accurate and efficient method for corn seed variety identification.
• This approach offers a valuable tool for rapid, non-destructive classification in smart agriculture.
• The innovative module combination method presents a new option for diverse agricultural applications.