An image-based system to preliminary assess the quality of grape harvest batches on arrival at the winery

•Our aim is to discern between grapes, grape stems and other objects.•With a V-shaped histogram, thresholding can be automatized.•Our algorithm is based on spectral and spatial data filtering.•Direct sunlight produces shadows and strong light spots, failing the identification.•The system works outdoors and is more stable when adding artificial light.

In this paper we present a hardware system and related computer processes to assess a preliminary quality control of grape harvest batches by automatically detecting, classifying and quantifying grape stems and other objects from images acquired on arrival at the winery, from which a quality grading is given. A non-invasive methodology based on computer vision is implemented, and the developed approach is based on the combination of a spectral and spatial analysis, where the inspection is performed in a small region of the near infrared (from 1000 to 1100 nm range). A non-expensive optical system was designed, which is composed of two high resolution CCD cameras, a long pass filter, some halogen light sources, related electronics and a standard PC. The performance of the optical system and implemented computer vision-based algorithms were tested in a laboratory setup for repeatability and overall accuracy while operating outdoors and over a whole working day, with varying illumination conditions. The system response presented small variations when estimating the occurrences of grape stems or other objects in the scene under analysis. The best results – lowest standard deviations and lowest root mean square errors – were obtained for the system working with added artificial lights and for the analysis being performed in the green channel. The system was also installed at a winery, where operators tested it for its good performance in a real scenario. The system worked as expected and the outcomes show that it is a robust and intuitive inspection system.