Seed quality analysis using computed tomography

A system comprising:

• a data processing apparatus including at least one hardware processor;
• and
• a non-transitory computer-readable medium encoding instructions configured to cause the data processing apparatus to perform operations comprising:
• obtaining computed tomography data representing a germination towel comprising seedlings;
• detecting likely seeds and candidate shoots and roots in the computed tomography data representing the germination towel;
• forming segments from the candidate shoots and roots, the segments being specified in the computed tomography data using segmentation masks or bounding boxes;
• grouping the segments across a total volume of the computed tomography data into likely shoots and roots in accordance with relative proximity and alignment of the segments;
• assembling the likely seeds and the likely shoots and roots into likely seedlings;
• classifying the germination towel using a classifier trained to use at least the likely seedlings, as input;
• and
• providing a classification result from the classifying for the germination towel.

A system comprising:

• a data processing apparatus including at least one hardware processor;
• and
• a non-transitory computer-readable medium encoding instructions configured to cause the data processing apparatus to perform operations comprising:
• obtaining computed tomography data representing a germination towel comprising seedlings;
• detecting likely seeds and candidate shoots and roots in the computed tomography data representing the germination towel;
• forming segments from the candidate shoots and roots;
• grouping the segments across a total volume of the computed tomography data into likely shoots and roots in accordance with relative proximity and alignment of the segments;
• assembling the likely seeds and the likely shoots and roots into likely seedlings;
• classifying the germination towel using a classifier trained to use at least the likely seedlings, as input;
• and
• providing a classification result from the classifying for the germination towel, wherein the detecting comprises:
• identifying centers of the candidate shoots and roots by processing the computed tomography data using a blob detection algorithm,
• wherein the identifying comprises:
• convolving data derived from the computed tomography data using a kernel function to generate convolved computed tomography data, wherein a shape of the kernel function is determined in accordance with a dimensional property associated with the shoots and roots of the seedlings;
• computing a Hessian matrix for each voxel of the data derived from the computed tomography data;
• computing a Laplacian for each voxel of the data derived from the computed tomography data using the Hessian matrix;
• selecting voxels of interest using at least the Laplacian for each voxel;
• and
• determining the centers of the candidate shoots and roots using the voxels of interest.

A non-transitory computer-readable medium encoding instructions operable to cause data processing apparatus to perform operations comprising:

• obtaining computed tomography data representing a germination towel comprising seedlings;
• detecting likely seeds and candidate shoots and roots in the computed tomography data representing the germination towel;
• forming segments from the candidate shoots and roots;
• grouping the segments across a total volume of the computed tomography data into likely shoots and roots in accordance with relative proximity and alignment of the segments;
• assembling the likely seeds and the likely shoots and roots into likely seedlings;
• classifying the germination towel using a classifier trained to use at least the likely seedlings, as input;
• and
• providing a classification result from the classifying for the germination towel,
• wherein the detecting comprises:
• convolving data derived from the computed tomography data using a kernel function to generate convolved computed tomography data, wherein a shape of the kernel function is determined in accordance with a dimensional property associated with the shoots and roots of the seedlings;
• and
• identifying centers of the candidate shoots and roots using the convolved computed tomography data.