Estimation of grapevine lateral leaf area (LLA) from RGB imagery using color thresholds and automatic k-means image segmentation

Lateral leaf area (LLA) is an alternative method that can be used to evaluate canopy size and leaf area index (LAI). LLA estimation requires accurate image segmentation since conditions such as shadows, light, or soil interference need to be excluded to increase accuracy. In this study, a field experiment was set up to evaluate LLA determinations using different color spaces transformations together with manual color threshold and automated k-means segmentation. In the field, RGB images were obtained of the east side of the canopies at mid-morning using a white panel background to avoid the interface of subsequent rows. Four color spaces: RGB (red, green, blue), HSV (hue, saturation, value), YCbCr (brightness, Cb, Cr) and CIELab (luminance, green to red, blue to yellow) were evaluated for a color thresholding process. Also, an automatic method using k-means classification was evaluated. Each image was first subjected to a detailed manual image segmentation process to create a reference LLA-image which represented the true LLA value where only the leaves of the canopy are present. Manual and automatic methods were compared against this reference LLA-image using a MATLAB script to obtain the accuracy of the methods (% of agreement). Results show that the automatic k-means method presented good performance for low and medium vigour vines. Among the color spaces tested, HSV was the most promising for LLA estimation, regardless of the density of the canopy with a consistent agreement of around 80 to 90%. The methodology presented is accurate, inexpensive and can be applied to seasonal LLA-image data to evaluate spatial and temporal variability of canopy size.