Developing a mobile App to estimate grape volume and colour for harvest

Recently, mobile Apps using image processing and machine learning have gained popularity for various real-life agricultural applications due to its low cost, portability and easy to use nature. This poster will present an Android App development procedure to extract grape berry volume and colour features to predict a suitable harvest window according to preferred wine style for large-scale vineyard management. This App detects and segments berries within images based on colour and shape. In this process, berries are detected and segmented individually using Circular Hough Transform (CHT) technique. This technique provides the radius of each berry, and is used to determine berry volume. In practice, automatic brightness control and image correction operations prevent a smartphone camera from determining correct colour information. Consequently, an illumination correction algorithm was developed for determining accurate colour information to cope with variable ambient light conditions associated with field derived images. For displaying consistent colour, gamma-correction requires an image with standard red, green and blue (sRGB) colour space. The sRGB gamma-corrected image is converted to the CIE XYZ colour space for illumination correction algorithm. Then, the hue colour angle (HCA) is determined for each pixel, which is a measure of berry colour. Depending on the HCA distributions, the approximate colour of the berry is determined. This process can determine correct HCA under different weather condition (i.e., sunny, cloudy or any shade). These HCA and volume information will assist to predict harvesting date and recommendations to grape growers and winemakers.