Early detection of sugarcane diseases is important for effective management strategy. However, it is a challenging task as many diseases can hardly be detected in the early stage of infection due to a lack of human-eye-detectable symptoms. For enhanced detection, hyperspectral imaging technologies can capture light from a broad spectral range which often includes part of the invisible spectral range (e.g., near-infrared), and thereby can potentially capture the features for early detection of sugarcane diseases before the development of visible symptoms. Moreover, deep neural networks are well-developed frameworks from the field of deep learning that can automatically extract features for association with specific disease-related spectra. In this research, early detection of sugarcane smut and sugarcane mosaic diseases was achieved using a combined hyperspectral imaging and deep learning approach that was subsequently applied to track disease development over a five-month trial period. Additionally, the first large-scale high-resolution hyperspectral image dataset of inoculated and healthy (control) sugarcane plants was created and released. Subsequently, experimental results indicated that hyperspectral images containing features that can be used for the early detection of the two target sugarcane diseases were extracted by deep neural network models efficiently. The detection accuracy for both diseases was above 90 % before the appearance of visible symptoms.