IMAGE PROCESSING TECHNIQUES FOR AUTOMATED COFFEE QUALITY EVALUATION

The subject of this thesis is the research and development of automated techniques for automated coffe quality evaluaton. intended to substitute the present subjective evaluations carried out by quality control experts. Classic evaluations take into account factors such as crema volume, persistence, foaminess, and color.. The aim of this work is to standardize and speed up the assessment process from one coffee blend to another and among different machines by obtaining objective, reproducible measurements. The proposed system is based on image processing technology and specifically tailored tools that maximize and automate image taking. The volume of coffee and crema is the initial parameter addressed. Current practice relies on milliliters graduated beakers and visual estimation, which are subjective and are not sensitive to coffee's behavior in a real cup. To combat these issues, two common cup sizes (cappuccino and espresso) were used, along with an image-processing tool that estimates liquid volume from side images. A 3D-printed object was designed to ensure regular imaging: it has a camera mount, slide base for varying cups, and a calibration panel to convert pixels to centimeters to ensure consistency even with changing zoom and lighting. For crema volume measurement, a more advanced tool is needed to record crema collapse shortly after extraction. Since the crema settles irregularly and leaves color trails which are deceptive to height measurements, the method involves chromatic averaging at several points, not including the top meniscus. This allows improved assessment of crema persistence. Experimentally, crema persistence has traditionally been measured with an aluminium mesh 30 seconds after extraction, an operator-dependent method that is prone to variability and utilisation of non-representative vessels. The new technique involves acquiring sequential images over a two-minute timescale to determine the percent reduction in crema over time. Experiments on multiple machines demonstrated crema degradation that was evident, justifying the merit of the conceptualised technique. Further, the upper surface view of crema,used in the purpose of assessing foaminess and white spots,has been resolved. Computer vision modeling was used to assess bubble size and distribution. A set of tagged data, collected across a range of different blends and lighting configurations, facilitated the training of a strong model to identify and quantify bubble density, filtering out reflections and noise bubbles from the glass. The white spot percentage, which has a propensity to appear at the extraction point of the coffee and is usually not desirable, is now visually evaluated as acceptable or otherwise. An additional device calculates the exact area of the spot affected based on pixel thresholds set by testers, converting the image to a binary mask to determine the spot percentage. For easy use, a graphical user interface (GUI) was created so that the user can load an image, choose the cup type, and see the processed output. Automated techniques were compared with conventional ones, demonstrating sound consistency in results. The study suggests a trustworthy, unbiased system that reduces human error and minimizes the time required to evaluate the quality of espresso in a cup.