Multi-Spectral Image Synthesis for Crop/Weed Segmentation in Precision Farming

An effective perception system is a fundamental component for farming robots, as it enables them to properly perceive the surrounding environment and carry out targeted operations. The most recent methods make use of state-of-the-art machine learning techniques learn valid model target task. However, those need large amount labeled data training. A approach deal with this issue augmentation through Generative Adversarial Networks (GANs), where entire synthetic scenes are added training data, thus enlarging diversifying their informative content. In work, we propose an alternative solution respect common methods, applying problem crop/weed segmentation in precision farming. Starting from real images, create semi-artificial samples by replacing relevant object classes (i.e., crop weeds) synthesized counterparts. To do that, employ conditional GAN (cGAN), generative trained conditioning shape generated object. Moreover, addition RGB take into account also near-infrared (NIR) information, generating four channel multi-spectral images. Quantitative experiments, carried on three publicly available datasets, show that (i) our capable realistic images plants (ii) usage such process improves performance semantic convolutional networks.