a feature extraction method based on deep learning using hyperspectralimaing for the evaluation of oilseed repe canopy nitrogencontent grades

摘要

Efficient and accurate processing of large quantities of hyperspectral image (HSI) data is a challenging for the Non-destructive evaluation of nitrogen content in oilseed rape (Brassica napits L.). This research aimed to develop a fast nitrogen Contents grades classification method for Oilseed rape canopy by employing a deep learning algorithm named stacked auto-encoders (SAEs). In this study, 5 nitrogen degrees were applied for the oilseed rape samples and hyperspectral image data was acquired under3 camera angles (0°, 15° and 25°) during the two growing stages before flowering. The algorithms SAEs was introduced for the data dimensional reduction and feature extraction from hyperspectral images, then multiple classification models were appliedfor the feature testing and validation within the feature data under different camera angles with different feature units. Our result shows the data under 25° shows the best classification result, besides, SAEs shows promising performance in HSI big data processing in phenotyping tasks.