multispectral images from the orchard, detecting HLB citrus disease

The dataset contains a collection of  multispectral images constituted of 14 VisNIR bands. Images were collected in-situ in brazillian citrus  commercial orchard, in the state of Sao Paulo. They describe Large portions of canopy approximately covering 3 m². It was designed to study HLB disease in citrus,specifically its detection in orchard with computer vision. The datset was formated to be used input for machine leaerning and deep-learning algorihms, as a classification task. The dataset is dived into two plots. "PlotA" acquired between August 15th and 18th 2023, in a 9ha of sweet orange Pera Rio (Citrus sinesis (L.) Osbeck), which was three years old. It contains 1,297 multispectral images. "Plot B"  acquired from Novemebr 18 to 20, 2024, containing  1681 multispectral images. Images are splitted into two classes: "HLB-symptomaric" and " non-HLB symptomatic" Two formats are proposed. "Raw", where each multispectral image is a folder containing 14 files in "tiff" format, i.e. one monocghromaic image per sepctral band. "Datacube",  in which each multispectral image is an hdf5 file containing a cube channel x width x height to be extracted as a tensor. e.g. "PlotA_datacubes_hlb.zip" is a folder containing all multispectral images with hlb symtoms in plot A, each as ".h5" file. For convinience and to reduce folder size under 50 Gb, folders can be splitted into _part_ . Images were all acquired with a Sony Alpha 7R2 Sextuple caméra. The caméra has 6 lenses each associated to a  spectral filter. Spectral channels (nm) 405, 430, 450, 490, 525, 550, 560, 570, 630, 650, 685, 710*, 735*, 850# Bandwidth 25 nm; *15 nm; #10 nm Focus Manual 0.5 m to infinity F-number (Aperture) Fixed 5.6 Field of view (FOV) Diagonal 45.9°; Horizontal 35.0°; Vertical 26.6° Focal size 9600 x 6376 pixels Focal length 21.8 mm final image dimension: 2550 x 2560-pixels This acquisition methods suffer from strong parallax effects and might require regitration operations. example of codes to format image, read dataset, register images, train and validate deep learning models can be found at: https://github.com/florent-abdelghafour/MSI-DeepLearning