Zoek in de site...

anomaly detection in HSI using data inner polygon

Supervisor: dr. Mahdiyeh Ghaffari

  • HyperSpectral Images (HSI)

Hyperspectral imaging measures the spatial and spectral features of any object at wavelengths ranging from ultraviolet through long infrared, including the visible spectrum. A Hyperspectral image (HSI) can include dozens or hundreds of channels unlike RGB images, which use only three types of sensors sensitive to the red, green, and blue portions of the visible spectrum. Hyperspectral images have more information than RGB images. Two identical objects in an RGB image can differentiate in HSI, due to different characteristic properties.

  • Anomaly detection

Anomaly detection is an important branch in chemometrics/machine learning and has important applications in computer vision and data mining. The purpose of anomaly detection is to identify a part of the data that is not normal or is significantly different from the rest. Anomaly detection in HSI refers to the identification of pixels whose spectral characteristics in an image are significantly different from global background pixels (it can be a pixel or multiple pixels). In fruit quality control using HSI, anomaly detection can be used to distinguish fresh fruit from damaged ones. Figure 1 presents two strawberries that are fresh (a) and fungi infected (b). This can be an example to confirm/reject the healthiness of and strawberry. In the current example, RGB is enough to distinguish. However, in the case of a minor number of pixels, it will be difficult and HSI is needed. In addition, HSI will help to characterize the type of impurity as well.

Figure 1. fresh (left) and fungi-infected (right) strawberries.

In this internship, you will have a look at the literature on anomaly detection methods first. Later you will get familiar with Principal Component Analysis (PCA) and data inner polygon in PCA space. Next, you will work on the application of data inner polygon in anomaly detection in HSIs. Finally, you will use the modified/generated algorithm in real cases/HSIs. All these analyses will be automated in MATLAB.

This internship will help to increase your:

  • Ability to image pre-processing/processing
  • Programming experience (MATLAB)
  • Ability to critically review scientific literature
  • Ability in big data analysis/visualization

The length and objectives of this internship can be adapted to bachelor and master internships in consultation with the supervisor at the start of the internship.