Researchers at the College of Agronomy, Hunan Agricultural University are applying hyperspectral imaging as a powerful tool to improve rapeseed growth, yield, and quality. Their work focuses on addressing key challenges in managing rapeseed (Brassica napus L.) production, an important oilseed crop cultivated worldwide.
One of the most critical phases for maximizing seed output is the pod-filling stage, when around two-thirds of the plant’s dry matter is produced through photosynthesis within the pod. During the green-ripe and yellow-ripe stages, the pod’s ability to perform photosynthesis is closely tied to its water content, making accurate monitoring during these periods essential.
A key indicator of pod development is the water content of the pericarp, the outer skin, or wall of the pod. Measuring pericarp water content provides valuable insight into the pod’s overall moisture level and maturity — both of which affect how vulnerable the pod is to shattering. As pods mature and dry out, they become more prone to splitting open, leading to significant harvest losses—sometimes as much as 20%.
Being able to accurately track pod pericarp water content is crucial for optimizing harvest timing and protecting yield. However, traditional methods, like drying and weighing the pod pericarp, are slow, destructive, and not practical for large-scale monitoring. Farmers have often relied instead on visual cues, such as widespread yellowing of pods or aiming for a specific moisture range, but these methods are subjective and inconsistent.
Hyperspectral imaging offers a promising alternative. It enables rapid, non-destructive assessment of pod pericarp water content and maturity, supporting smarter, more efficient crop management practices. Through this technology, researchers are helping to modernize rapeseed production, improving yields and maintaining crop quality.
Study Design and Data Collection
reflectance samples (right). The five blue square-shaped points are the first round of random measurements, with the average value taken as the first spectral reflectance value. The red, green, pink, and orange represent the second, third, fourth, and fifth round of selection, respectively. The average of the five rounds is averaged as the final spectral reflectance value.
For this experiment, researchers used the Fengyou 520 rapeseed cultivar developed by the Hunan Crop Research Institute. Pod samples were collected at three critical stages of development: green ripe, yellow ripe, and full ripe. To ensure consistency, sampling was conducted under stable weather conditions, with pods selected from different sections of the plants. Immediately after collection, the pods were analyzed in the laboratory using hyperspectral imaging and water content measurements.
An SOC-710 hyperspectral camera was used to capture hyperspectral imaging data, measurements were conducted in a darkroom chamber coated with diffuse reflective material. The setup included a 70 W halogen light source positioned at a 45-degree angle and a ventilation and cooling system. The camera was mounted vertically above the sample at a distance of 300 mm. Spectral data processing was carried out using the Spectral Radiance Analysis Toolkit and ENVI 5.3 software.
A high-precision TP300D electronic balance and a 101-type drying oven were used to measure pod pericarp water content. Before drying, seeds were removed from each pod to avoid oil volatilization affecting results. Fresh and dry weights were recorded to calculate water content. A total of 216 samples were collected, along with matching hyperspectral and RGB images. Results showed similar water content in the green and yellow stages, but a clear drop and more variability at full maturity.
Hyperspectral Data
To improve modeling efficiency and avoid redundant data, researchers focused on four main types of features from hyperspectral images: image color, image texture, spectral edge parameters, and spectral indices.
- Image Color: After removing background noise, 98 color features were extracted from the pods.
- Image Texture: Texture features such as contrast, energy, correlation, entropy, third-order moments, brightness, consistency, homogeneity, and smoothness were captured to describe surface patterns in the pod images.
- Spectral Three-Edge Parameters: By analyzing the red (680–760 nm), blue (490–530 nm), and yellow (560–640 nm) spectral edges—areas sensitive to plant health and biomass.
- Spectral Indices: The crop spectral index is the feature obtained by combining spectral reflectance from sensitive wavelength regions, which is very useful for assessing and monitoring crop growth. Ratio spectral index (RSI), normalized difference spectral index (NDSI), and difference spectral index (DSI), which are most important for green crops were used for this study. In addition, four other key spectral indices related to plant water content, namely plant senescence reflectance index (PSRI), ripening stage of pod maturity index (RSMI), structure-independent pigment index (SIPI), and modified normalized difference red edge index also employed.
To estimate rapeseed pod water content and recognize maturity levels, researchers used a feature-level information fusion strategy. They selected the top three most relevant features from each category (color, texture, spectral indices, and edge parameters), combining them within and across categories. These fused features were used to train models using random decision forests (RDF), a supervised learning method known for its flexibility and robustness. The RDF models were compared with other approaches like linear regression, support vector machines, and neural networks to evaluate performance.
Conclusion
This study used hyperspectral imaging to develop models for estimating rapeseed pod pericarp water content and recognizing pod maturity. By fusing features such as color, texture, spectral edge parameters, and spectral indices, researchers found that combining features across categories improved model accuracy and stability. The best results were achieved using a fusion of spectral indices and three-edge parameters. Random decision forests performed best for water content estimation, while support vector machines were more effective for maturity recognition.
Although few studies have addressed rapeseed pod maturity or water content estimation, this study evaluated two spectral indices: the Canola Pod Maturity Index (CPMI) and the Rape Siliques Maturity Index (RSMI). While CPMI performed well for canola, it showed a poor correlation with the water content of the rapeseed pods used in this study. In contrast, RSMI—developed for Chinese rapeseed cultivars—showed a strong correlation, highlighting how regional differences in plant varieties can affect the effectiveness of spectral indices.
Learn More
To read the full article see:
Zhao, Z.; Liao, G. Imaging Hyperspectral Feature Fusion for Estimation of Rapeseed Pod’s Water Content and Recognition of Pod’s Maturity Level. Mathematics 2024, 12, 1693. https://doi.org/10.3390/math12111693