Prediction of Grapevine Leaf Identity with DenseNet121

This work looks into if DenseNet121 lets one achieve grapevine leaf recognition. The model consists of four dense blocks with 58 layers, a first convolutional layer, average, and maximum among other pooling layers. Comprising 400 training and 100 validation photos, the 500 images in the dataset are grouped into five categories. Important hyperparameters include a 256 by 256 image size, a 0.001 learning rate, a 0.45 dropout rate, and a 32-batch size. With a 92% accuracy, the model dropped 0.377 on the validation set going through up to 100 episodes. Highly interesting for automated agricultural uses, the suggested approach solves the grapevine leaf categorization problem.