Feature Extraction, Pattern Recognition and Classification in X-ray Image Data

Excellence of food products highly depends on the quality checks at different stages while preparation and processing at industry. With the evolution of technology, traditional methods are being put back and state of the art equipment taking the position. Being fast, efficient and automatic, computers and machines are potentially replacing the human deployment in the food industry. One of the early stages of food preparation is the ingredient evaluation on feeding belt. This is still carried out mostly by humans; however efforts have been made for the development of such system which is capable to inspect the ingredient quality in an automatic way. The research work involves developing and estimating such an arrangement which provide the quality information of ingredient without human deployment. X-ray imaging was employed for internal analysis of ingredients: pine, pistachio and hazelnuts. A captured x-ray image containing few non-overlapping ingredients was analyzed using image processing techniques to develop a method for automatic detection and extraction of independent ingredient. Individual ingredient image samples were further analyzed to calculate the strong features on the global as well as local level. A number of features including statistical, texture and moment invariant properties were extracted from each image sample and were organized in diverse combinations to be utilized further. Different databases have different percentage of representation for healthy and unhealthy nuts so correspondingly several classification techniques were exercised including logistic regression, artificial neural network, anomaly detection and support vector machines. In addition to accuracy, the percentage of correct recognition unhealthy ingredients was observed which is vital. Concluding fine classification accuracy was observed with comparatively better false positive rate than related studies