Related Experiment Video
Updated: Apr 14, 2026
08:25
Using Flatbed Scanners to Collect High-resolution Time-lapsed Images of the Arabidopsis Root Gravitropic Response
Published on: January 25, 2014
12.9K
Detection of Pumpkin Seed Plumpness Using Terahertz Imaging Enhanced With Compressed Sensing and Super-Resolution.
Bin Li1, Yiying Yang1, Jinli Yang2
1Institute of Intelligent Mechanical and Electrical Equipment Innovation, East China Jiaotong University, Nanchang, China.
Journal of Food Science
|April 13, 2026
Summary
This study introduces a non-destructive method using terahertz imaging and AI to assess pumpkin seed plumpness, improving quality detection accuracy and efficiency for agricultural applications.
Area of Science:
- Agricultural Science
- Non-destructive Testing
- Image Processing
Background:
- Pumpkin seed quality, crucial for viability and market value, is often assessed destructively.
- Traditional methods for evaluating internal seed quality are inefficient and time-consuming.
- Internal plumpness is a key indicator of pumpkin seed quality, affected by storage and processing.
Purpose of the Study:
- To develop and validate a non-destructive method for assessing pumpkin seed internal quality.
- To evaluate the effectiveness of terahertz (THz) time-domain imaging combined with compressed sensing (CS) and advanced AI for super-resolution imaging.
- To optimize reconstruction algorithms and measurement matrices for accurate THz image reconstruction.
Main Methods:
- Utilized a non-destructive terahertz (THz) time-domain imaging system.
- Integrated compressed sensing (CS) with the Real-World Enhanced Super-Resolution Generative Adversarial Network (Real-ESRGAN) for image reconstruction and enhancement.
- Compared five measurement matrices (e.g., Gauss Matrix) and five reconstruction algorithms (e.g., ADMM_TV, BP, SWOMP), selecting GaussMtx and ADMM_TV for optimal performance.
Main Results:
- The combination of Gauss Matrix (GaussMtx) and Alternating Direction Method of Multipliers-Total Variation (ADMM_TV) yielded the best image reconstruction quality (PSNR, NMSE, SSIM).
- Real-ESRGAN significantly improved the edge sharpness and detail of the reconstructed THz images.
- Achieved a low fullness error of 2.23% and an average detection error of 3.37% on the verification set.
Conclusions:
- The integrated THz imaging, CS, ADMM_TV reconstruction, and Real-ESRGAN super-resolution processing offers an efficient and accurate non-destructive method for pumpkin seed quality detection.
- This technology enables rapid, non-damaging assessment of seed plumpness, suitable for automated quality control in the seed industry.
- The findings contribute to reducing agricultural waste and enhancing the value of pumpkin seeds for consumers and planting.
Keywords:
Real‐ESRGANcompressed sensingplumpnesspumpkin seedsterahertz (THz) time‐domain imaging system
