| Machine vision solutions are becoming a standard for quality inspection in several manufacturing industries. In the processed-food industry where the appearance attributes of the product are essential to customer's satisfaction, visual inspection can be reliably achieved with machine vision. But such systems often involve the extraction of a larger number of features than those actually needed to ensure proper quality control, making the process less efficient and difficult to tune. This work experiments with several machine learning techniques in order to automate the initial tuning of a real-time vision-based food inspection system or to improve its performance. The impact of feature selection techniques on machine learning is also assessed. Identifying and removing as much irrelevant and redundant information as possible for a given learning scheme reduces the dimensionality of the data and allows classification algorithms to operate faster. In some cases, accuracy on classification can even be improved. The effect of filter-based and wrapper-based feature selectors is experimentally evaluated on different bakery products to identify the best performing approaches when combined with three fundamentally different machine learning strategies. |
