Digital image processing using convolutional neural networks

An image processing server comprising:

• a processor;
• a memory storing machine readable instructions that are to cause the processor to:
• create, by a training circuit, a training set from images of damaged objects;
• select a candidate architecture and candidate parameters for a convolutional neural network (CNN) to classify the extent of damage for the object in the image through an iterative process, wherein the iterative process comprises:
• selecting, by a model builder, the candidate architecture from a plurality of candidate architectures;
• selecting, by the model builder, the candidate parameters for the selected candidate architecture;
• building, by the model builder, an intermediate CNN using the training set;
• evaluating, by the validation circuit, performance of the intermediate CNN on a validation set, and
• repeating the iterative process until it is determined that a predetermined number of intermediate CNNs meet a validation threshold, wherein each intermediate CNN has different values for the selected candidate parameters;
• creating, by the validation circuit, an ensemble of intermediate CNNs from the predetermined number of intermediate CNNs;
• and
• classify, by a classifier, the extent of damage for the object in each image in the validation set, wherein to classify is to aggregate predictions from the ensemble of intermediate CNNs.

A non-transitory computer readable medium to process digital images, including machine readable instructions executable by a processor to:

• select a candidate architecture and candidate parameters for a plurality of convolutional neural networks (CNNs) to classify the extent of damage for the object in the image;
• determine that a predetermined number of CNNs meet a validation threshold, wherein each CNN has different values for the selected candidate parameters;
• select an ensemble of CNNs from the predetermined number of CNNs;
• aggregate predictions from the ensemble of CNNs;
• and
• classify the extent of damage for the object in the image.