Methods of image acquisition and software development for leaf area measurements in pastures

The development of software for automated image processing, with particular emphasis on agricultural monitoring applications, has increased in the last years. Leaf area measurements are important for several crops and pastures, once the leaf area index is related with growth and photosynthesis rates, being an essential parameter of process-based models of vegetation. The processing algorithms implemented in most software for leaf area measurements currently available require the determination of leaf dimensions. Hence, the automated processing of samples with multiple leaves is still limited, frequently requiring manual pre-processing of the images. In order to develop a new software system aimed at processing images of samples composed of multiple leaves without any requirements of manual pre-processing of the images, the USP-Leaf software was developed. The software’s performance was compared by using different devices for image acquisition: a semiprofessional digital camera (Sony, resolution of 12 megapixels); a mobile phone (Lenovo k5, resolution of 5 megapixels) and a desktop scanner (HP B20, 300 dpi corresponding to 3.508 × 2.480 pixels). The results were validated by comparing the values obtained with the standard method (electronic planimeter model Li-Cor 3100). The experiment was carried out at the Faculty of Animal Science and Food Engineering (FZEA), University of São Paulo, Pirassununga, SP, Brazil. The leaf samples were obtained from Brachiaria decumbens Stapf. cv. Basilisk pastures. A total of 20 samples comprising 15 leaves were collected, from which the images were acquired with each device. Edge detection, filtering and thresholding algorithms were applied to identify the leaf section of the image against the background. Considering the leaf area measured with the electronic planimeter, the relative error rate of the software’s estimates was lower than 7%, being highest when the scanner was used and lower with the digital camera. Pearson’s correlation coefficients were higher than 95%, regardless of the device used for image capturing, indicating that the software was able to provide accurate estimates of leaf area. The linear regression equation associated with the estimated leaf area using the mobile phone showed the highest values for the intercept and the higher standard error associated with this parameter (2.9 ± 5.69), despite showing a slope close to 1 (1.0 ± 0.07). The leaf area estimates were close to the standard method, showing that the software’s performance was not affected by the device used for image acquisition.