Detection of Fungal Disease in Citrus Fruit Based on Hyperspectral Imaging

The dataset is here.

Dataset

The dataset can be obtained here. It contains recordings of:

• Healthy citrus 200 images
• Phytophthora syringae 185 images
• Phytophthora citricola 210 images
• Phytophthora citrophthora 205 images

​ Fig. 1.RGB images of three citrus diseases synthesized from visible light spectra.

Relevant information about hyperspectral imaging devices：

• The portable snapshot hyperspectral imaging system used in this study is composed of Specim FX 10e (Spectral Imaging.Ltd,Finland), a dark box, and a computer installed with SpecView data collection software. The camera is configured to capture images with dimensions of 1024×1024 pixels. Each hyperspectral image has 224 channels, with a spectral resolution of 5.5 nm, a spectral sampling interval of 2.7 nm, and a spectral range from 400 to 1000 nanometers, maintaining the visible light (VIS) range plus a lower near-infrared (NIR) range.

Requirements

• Python 3.9.13
• PyTorch 1.12.0
• visdom
• Download the data set to a local folder

Our model

This is the official implementation of the network, based on PyTorch.

The code is divided into subfolders, which correspond to the use cases:

• checkpoint contains the training process of all tasks. Here, you can find various pre-trained models, including the training results for RGB, raw spectral images, and dimensionality-reduced spectral images.

• dataset stores the dataset. Please extract the downloaded data into this directory.

• Model contains the implementation of various models.

• train.py contains the training code, where you need to manually open the annotation selection model. The default model is our_model.

• validAccResult.py Validate the model effect through the test set. Replace the saved model with the corresponding position in the code.The default model is Our_model_CARS.pth.