RhodaNet: A novel deep learning architecture for Rose disease classification

Technology adoption in agriculture has creatively solved and eased many farming problems. Home gardeners grow various plants throughout the year. Many of them show interest in growing roses at their houses. In a survey conducted with home gardeners 70% of them have roses in their garden and 80% of them added that their plants get infected often. Early and accurate diagnosis of the diseases may reduce the likelihood that the plant will suffer further harm and spread. The substantial advancements in deep learning have created the opportunity to improve the coordination and accuracy of the system for identifying plant diseases. An improvised ResNet architecture-RhodaNet is proposed to identify the rose disease at an early stage. RhodaNet architecture uses concatenation in stacked layers preserving both spatial and channel information even where input and output have different channel sizes or feature representations. RoseNet dataset from Mendeley data with 2993 images was considered for the study with 5 diseases Black spot, Downy Mildew Powdery Mildew. Mosaic, Botrytis Blight. Our proposed model gives an accuracy of 96% whereas ResNet and DenseNet gives 91.56% and 93.50% respectively. RhodaAPP predicts the type of disease and its remedy serves as an appropriate solution for home gardeners.