Search

The accurate phenotyping of the external quality attributes of potato tubers is important in potato breeding. Currently, the assessment of potato tuber shape, together with eye density and depth, are based on subjective naked eye visual evaluation. However, such a manual visual assessment makes it very difficult to reliably phenotype these and other important, more complicated, geometrical traits, such as shape uniformity. In this study, a 3D image analysis method has been developed for counting potato eyes and estimating eye depth based on an evaluation of the curvature of an acquired 3D point cloud. Six shape uniformity-related traits, together with their shape indices (SI), were measured for six potato varieties. These were collected from three field experiments designed initially to study the effects of variation in nitrogen (N), potassium (K) and compound fertilisers along with tuber mass, on all investigated external traits. We demonstrate that a 3D image analysis technique can estimate the number of potato eyes and their depth with a high degree of accuracy. In addition, three shape uniformity traits were identified as offering a better power discrimination between varieties. The preliminary experiment found potato tuber mass to significantly affect both the shape uniformity and eye count, while fertiliser treatments showed no effect on all traits except SI. However, further investigation with a larger sample size is required for confirmation.

Recently, effort has been made to apply deep learning to the detection of mesh saliency. However, one major barrier is to collect a large amount of vertex-level annotation as saliency ground truth for training the neural networks. Quite a few pilot studies showed that this task is quite difficult. In this work, we solve this problem by developing a novel network trained in a weakly supervised manner. The training is end-to-end and does not require any saliency ground truth but only the class membership of meshes. Our Classification-for-Saliency CNN (CfS-CNN) employs a multi-view setup and contains a newly designed two-channel structure which integrates view-based features of both classification and saliency. It essentially transfers knowledge from 3D object classification to mesh saliency. Our approach significantly outperforms the existing state-of-the-art methods according to extensive experimental results. Also, the CfS-CNN can be directly used for scene saliency. We showcase two novel applications based on scene saliency to demonstrate its utility.