Early detection of sugarcane diseases through hyperspectral imaging and deep learning

Early detection of sugarcane diseases is essential for the development of management strategies. Disease detection and monitoring in sugarcane rely heavily on visual inspection, which can be expensive and subjective. Many diseases can show symptoms only later in disease development, with the initial infection invisible to the naked eye. These challenges motivated us to explore technologies based on hyperspectral imaging and deep learning for detection of sugarcane diseases. Key steps included trial preparation, hyperspectral image dataset construction, hyperspectral image pre-processing, deep-learning model development, and validation of the effectiveness of the developed technology. Outcomes suggest that the proposed technology is promising for the detection of sugarcane smut and mosaic diseases, in some cases earlier than visual symptoms emerge. Diseases can be detected with high accuracy as early as 8 weeks after inoculation for smut and 2 weeks after inoculation for mosaic, before the symptoms become visible in week 10 and week 8 for smut and mosaic, respectively.