Application Case | FS-IQ Hyperspectral Camera Assists in Early Non-destructive Detection of Rice Bacterial Leaf Blight

Rice bacterial leaf blight is a major disease affecting rice yield and food security. Traditional field inspections struggle to identify the disease during the asymptomatic stage, and by the time lesions appear, the effectiveness of prevention and control is significantly reduced. Hyperspectral imaging, with its characteristic of combining images and spectra, can capture subtle physiological and biochemical changes caused by the disease, making it an important means for early diagnosis of plant diseases.

In a study oriented toward the early diagnosis of rice bacterial leaf blight, the scientific research team selected the FigSpec FS-IQ-VISNIR portable hyperspectral camera produced by CHNSpec to conduct data collection, providing a stable and reliable spectral data source for intelligent disease recognition.

I. Experimental Equipment and Data Collection

• Equipment Model: FigSpec FS-IQ-VISNIR Hyperspectral Camera.
• Spectral Range: 400-1000 nm, with a spectral resolution of 2.5 nm.
• Collection Conditions: Sunny daytime between 10:00–14:00; lens distance from the canopy was 60-80 cm; DN values were controlled at 3000-4000 by adjusting exposure time in real-time to reduce the impact of overexposure and noise.
• Experimental Objects: Rice leaf samples of three levels: healthy, mildly infected (asymptomatic stage), and severely infected.

The FS-IQ hyperspectral camera supports fast, non-contact imaging and can stably acquire leaf spectral information in both controlled environments and field scenarios, laying the data foundation for subsequent feature extraction and model training.

II. Data Preprocessing and Key Band Mining

The original hyperspectral data underwent dark current correction, white board correction, and Savitzky-Golay smoothing. After removing low signal-to-noise ratio bands at both ends, 243 high-quality bands were retained for modeling analysis.

The study used deep learning methods to filter out characteristic bands sensitive to bacterial leaf blight from the full spectrum, mainly concentrated in:

• Green Peak Region (520–550 nm): Related to changes in chlorophyll content.
• Red Edge Region (680–720 nm): Reflecting leaf cell structure and stress states.

Using only about 8% of the core bands can retain most of the discriminatory information, reducing data dimensionality while improving model operational efficiency and recognition stability.

III. Disease Recognition Effect and Application Value

In the classification and recognition task of bacterial leaf blight, model verification was conducted based on the spectral data obtained by FS-IQ:

• Using a small number of core bands as input, the classification accuracy reached over 96%, which was better than the direct input of the full spectrum.
• For scenarios with unbalanced samples, after expanding minority samples through generative methods, the overall performance of the model improved by 6%–13%.
• The band selection results were consistent with the laws of plant physiological changes, possessing good mechanistic interpretability.

The FS-IQ hyperspectral camera demonstrated the following adaptation advantages in this study:

• Rich bands and stable signal-to-noise ratio: Covering the key visible-near-infrared interval, it can capture weak spectral differences in the early stages of the disease.
• Portable and easy to use: Suitable for laboratory and in-situ field collection, adapting to crop phenotype analysis scenarios.
• Strong data compatibility: Output spectra can be directly connected to deep learning and machine learning workflows, supporting feature mining and model optimization.

IV. Summary

Targeting the early non-destructive detection of rice bacterial leaf blight, this case relied on the FS-IQ hyperspectral camera to obtain high-quality spectral data. Combined with intelligent algorithms, it achieved sensitive band extraction and precise disease recognition, providing a feasible technical path for early crop disease warning and precision prevention and control.

The CHNSpec FS-IQ series hyperspectral cameras, with stable imaging performance and a user-friendly operation experience, continue to serve scientific research and industrial scenarios such as smart agriculture, plant phenotypes, and food safety, helping users mine effective features from complex spectral information and promoting the development of detection technology toward non-destructive, efficient, and intelligent directions.

Product Recommendation: FS-IQ-VISNIR Portable Hyperspectral Camera

• Spectral Range: 400-1000nm
• Spectral Resolution: 2.5nm
• Image Resolution: 1920*1920
• Number of Spectral Channels: 1200