Common Mold Classification Using Histogram of Oriented Gradient and Convolutional Neural Network

Molds have an immense number of species with different classifications and characteristics. Each one of them has its own human health-related effects and conditions after being consumed. Some species are good, most of them are not. The accuracy of the conventional classification method depends on the experience, knowledge, and training of the microbiologist. This study focuses on determining the kind of mold species produced in a certain food, by a certain place and humidity. The only hardware requirements for this study are a laptop compatible with machine learning and neural network, a microscope, and a USB camera. The USB camera is attached to the eyepiece of the microscope, capturing microscopic mold images. This study requires many images of certain mold species to be recognized through machine learning. To maximize the efficiency of using feature extraction in image processing, a set of images needs to be gathered as training images for the neural network. The images to be loaded in machine learning should be at a microscopic level to attain accuracy improvements. A free software Scikit-learn machine learning in python is used in the study. With the use of both feature extraction methods, the Histogram of Oriented Gradient and Convolutional Neural Network have achieved an overall accuracy of 96.29% and precision of 91.67%.