An effective segmentation algorithm of apple watercore disease region using fully convolutional neural networks

The apple watercore disease region detection is crucial to apple disease analysis in life science. However, traditional methods cannot be applied to this situation, because the watercore region varies gradually according to the apple region, and lighting variations and the bruise also give many false alarms. Now there is no such a stable and accurate method to cope with this difficulty. In this paper, we propose an effective apple watercore disease region segmentation scheme, which is intended to extract comprehensive representations of image on the basis of multilayer convolutional neural networks. With the learnt hierarchical feature representations from millions of pixels, false alarms caused by lighting variation and bruise are also restrained. Experiments based on a large self-designed dataset of apple watercore disease images show encouraging results.