Guava (Psidium guajava) is one of the most important fruit crops in Brazil, being popular due to its availability in all seasons and rich nutritional and medicinal value. However, Costalimaita ferruginea, is one of the main pests in guava production and directly interferes with its productivity. Symptoms of attack are often determined by visual observation, which may lead to inadequate characterization of the damage caused by this pest. Areas of guava plants (blocks) and plants with four groups of leaves and five leaves per group and per plant (replicates) were considered. This study spectrally characterizes the spectral response of guava leaves under infestation by Costalimaita ferruginea (Coleoptera: Chrysomelidae).   The spectral signature of the leaves was determined using spectroradiometer. The spectral signatures obtained showed the uninfested, moderately infested, and severely infested. The uninfested leaves in showed lower reflectance between the 500-600 nm bands and higher reflectance between the 750-1000 nm bands. The infested leaves level 1 exhibited moderate reflectance between 500-600 nm and lower reflectance between 750-1000 nm. In contrast, the infested leaves level 2 had a higher reflectance between 500-600 nm and a lower reflectance between 750-1000 nm. The uninfested leaves showed lower reflectance in the visible spectrum and higher reflectance in the near-infrared (NIR) spectrum than the infested leaves levels 1 and 2. It is also noted that as the infestation becomes more severe, the reflectance in the visible increases and the NIR decreases, denoting the typical behavior of plants under biotic stress. PCA and linear regression confirmed the efficacy of hyperspectral reflectance in discriminating the damage levels of C. ferruginea. The data generated in this study can be integrated into hyperspectral bank for future applications in entomology, damage and pest monitoring.

Hyperspectral image acquisition:

The collection of hyperspectral remote sensing data was carried out based on the methods described by Pinto et al. (2020) and Iost Filho et al. (2022). A hyperspectral pushbroom camera, model PIKA L (RESONON Inc., Bozeman, MT, USA), was used with the following specifications: which collects 160 spectral bands in the range of 300 to 1000 nm (visible and NIR spectrum). The camera lens has a focal length of 35 mm (maximum aperture of F1.4) and has the following specifications: Firewire interface (IEEE 1394b); digital output (12-bit); 7 degree angular field of view; spectral resolution is 3.3 nm. The imaging system included a tower with four 15 W and 12 V LED lamps, a stable power source, and control software (PIKA L, Resonon Inc., Bozeman, MT, USA andcontrol software (PIKA L, Resonon Inc, Bozeman, MT, USA. A voltage stabilizer (Tripp-Lite, PR-7b, www.radioreference.com) was used to maintain stable illumination. A dark calibration was performed with the camera lid closed. Then, the lid was removed and a white Teflon plate (K-Mac Plastics) was used for white calibration on the sample holder.

Spectral extraction and data processing:

Hyperspectral data were manually extracted for each leaf using a standardized region of interest in SpectrononPro software. Average reflectance values were calculated to generate spectral curves. Although the sensor captured wavelengths from 325–1075 nm, only the 400–1000 nm range (281 channels) was analyzed to remove noisy edge bands with low signal-to-noise ratios. Reflectance values were normalized by dividing by 10,000. The study evaluated hyperspectral reflectance differences among non-infested, low-infested, and highly infested leaves. Data were processed using radiometric filters and analyzed with statistical and machine learning methods in SAS 9.4, including PROC MEANS, PROC ANOVA, and PROC GLM.

The classification of the different levels of damage was performed using Principal Component Analysis (PCA) with variances explained in the "Google Colab in Pandas DataFrame with Python package upgrades and inclusions (statsmodels and scipy.stats)". Data processing was performed using NumPy and Pandas libraries in Python 3. Results on line (https://numpy.org/cite/; https://pandas.pydata.org/about/citing.html) and via methods of Harris et al. (2020) and McKinney (2010). These steps seek to ensure reliable and reproducible results for the study of the spectral properties of the different levels of damage.