Intelligent Classification of Japonica Rice Growth Duration (GD) Based on CapsNets.

Area of Science:

• Agricultural Science
• Spectroscopy
• Artificial Intelligence

Background:

• Rice cultivation in China's cold regions, particularly Heilongjiang Province, faces risks from low temperatures and cold damage.
• Selecting rice varieties based on accumulated temperature zones (GD) is vital for preventing crop loss.
• Traditional methods for identifying rice GD are time-consuming and unreliable due to environmental factors.

Purpose of the Study:

• To develop an efficient, accurate, and intelligent method for identifying japonica rice growing degrees (GD).
• To address the limitations of traditional field investigations for rice GD identification in cold regions.

Main Methods:

• Utilized Raman spectroscopy to analyze seven japonica rice varieties from three accumulated temperature zones in Heilongjiang.
• Applied data preprocessing techniques including signal filtering, differencing, segmentation, and superposition for feature fusion and dimension transformation.
• Developed and implemented a capsule neural network (CapsNets) model with a convolutional layer and two capsule layers, utilizing a dynamic routing protocol.

Main Results:

• The CapsNets model achieved 89% accuracy on the training dataset and 93% accuracy on the test dataset after 160 training epochs.
• The combined approach of Raman spectroscopy and CapsNets demonstrated effectiveness in classifying and identifying rice GD.

Conclusions:

• Raman spectroscopy coupled with CapsNets presents a viable intelligent identification method for japonica rice GD.
• This approach offers a significant improvement over traditional methods for rice cultivation in cold climate zones.