Recent Technique for Crop Type Classification from Satellite Images

Remote sensing techniques have been developed to allow researchers to accurately classify large areas of crops at reduced cost. These techniques often relay on images of high temporal frequency in order to utilize the changes in crop reflectance characteristics due to their phonologies. However, image sets with high temporal frequency have relatively low spatial resolution and as the spatial resolution increases the temporal frequency of the image decreases. This trade-off makes it difficult to classify remotely sensed images in areas where the yield sizes are small. Currently, there are no good solutions to counter act the trade off in resolutions. This study aims to develop a new method for combining high spatial resolution data with low spatial resolution data in order to arrive at a superior crop type classification. We have built a rigorous mathematical frame work to accurately describe the problem to be solved and propose a theoretically optimal solution to solved it. This solution is then implemented and tested on both synthetic and real dataset as proof of concept. We show that by merging the data with different spatial and temporal resolutions, an improvement in accuracy up to a 20 can be achieved even if very few high spatial resolution images are available for a scene.