Image-to-Image Translation-Based Data Augmentation for Improving Crop/Weed Classification Models for Precision Agriculture Applications

Applications of deep-learning models in machine visions for crop/weed identification have remarkably upgraded the authenticity precise weed management. However, compelling data are required to obtain desired result from this highly data-driven operation. This study aims curtail effort needed prepare very large image datasets by creating artificial images maize (Zea mays) and four common weeds (i.e., Charlock, Fat Hen, Shepherd’s Purse, small-flowered Cranesbill) through conditional Generative Adversarial Networks (cGANs). The fidelity these synthetic was tested t-distributed stochastic neighbor embedding (t-SNE) visualization plots real each class. reliability method as a augmentation technique validated classification results based on transfer learning pre-defined convolutional neural network (CNN) architecture—the AlexNet; feature extraction came deepest pooling layer same network. Machine support vector (SVM) linear discriminant analysis (LDA) were trained using vectors. F1 scores model increased 0.97 0.99, when additionally supported an dataset. Similarly, case technique, F1-scores 0.93 0.96 SVM 0.94 LDA model. show that generative adversarial networks (GANs) can improve performance with added advantage reduced time manpower. Furthermore, it has demonstrated could be great tool applications agriculture.