Sausage Colony Detection Based on Hyperspectral Image Analysis via Machine Learning

Sausage is a kind of food that people often eat in our lives because it is cheap and delicious. However, in the process of sausage production and transportation, if the quality control is not strict, it will cause deterioration and decay. If people eat these sausages, they will be sick and even die. Existing detection methods, such as physical and chemical detection, need a strict testing environment, and the sensory detection is inaccurate and incomplete. Colonies number is a key quality measure of sausage. Therefore, we planed to study a method to achieve rapid, non-destructive and accurate detection of sausage colonies number. In this paper, a method for detecting sausage colonies number based on hyperspectral technique and image features was proposed. Firstly, the sausage hyperspectral image and spectral information were collected. Then, the N-Dimensional Visualizer was used to extract features from the hyperspectral image. Moreover, the characteristic samples of sausage surface colony information were obtained by studying the colony surface characteristic spectrum curve and the color statistics of sample graphics. At last, the colony spectrum library was established and the two feature data were fused via the machine learning algorithm Support Vector Machine (SVM) to classify sausage quality. According to the established surface colony characteristic model, the samples of validation set were detected, and the fast detection of sausage colony quality was realized. The experimental results showed that the pixel proportion of the colony information in the image was consistent with the actual colony information measured in the physical and chemical experiments, which proved that the method is feasible to study the colony on the sausage surface. The method in this study can also be further extended to the detection of other related meat products.