Related Experiment Video
Updated: Jun 1, 2025
11:37
RGB and Spectral Root Imaging for Plant Phenotyping and Physiological Research: Experimental Setup and Imaging Protocols
Published on: August 8, 2017
16.1K
Multispectral imaging-based detection of apple bruises using segmentation network and classification model
Yanru Fang1, Hongyi Bai1,2, Laijun Sun1,2
1College of Electronics and Engineering, Heilongjiang University, Harbin, China.
Journal of Food Science
|January 20, 2025
Summary
This study introduces a deep learning method using multispectral imaging to accurately detect apple bruise levels and timing. The approach significantly improves bruise detection and classification, offering a new tool for the fruit industry.
Area of Science:
- Agricultural Science
- Computer Vision
- Machine Learning
Background:
- Bruises negatively impact apple appearance, nutritional value, and marketability, leading to economic losses.
- Accurate and timely detection of bruising is essential for quality control in the apple industry.
Purpose of the Study:
- To develop and validate a novel method for precise detection of apple bruise levels and timing.
- To enhance the accuracy of bruise segmentation and classification using deep learning and multispectral imaging.
Main Methods:
- A self-designed multispectral imaging system was combined with an improved DeepLabV3+ model for bruise segmentation.
- Depthwise separable convolution, efficient channel attention, and focal loss were employed to enhance segmentation accuracy.
- Spectral data from bruised regions were analyzed using improved DenseNet121 for bruise level and time identification, incorporating cosine annealing and attention mechanisms.
Main Results:
- The improved DeepLabV3+ achieved high intersection over union (IoU) scores (up to 95.5%) and F1-scores (up to 97.5%) for bruise segmentation.
- The enhanced DenseNet121 model demonstrated superior performance in identifying bruise levels (up to 99.5% accuracy) and bruise time (up to 99.3% accuracy).
Conclusions:
- The proposed deep learning-based multispectral imaging method offers a highly accurate and effective solution for detecting apple bruise levels and timing.
- This technology has the potential to significantly reduce economic losses and improve quality control in apple production and supply chains.
