Image uncertainty prediction method and device, equipment and storage medium

Abstract

The invention relates to an image uncertainty prediction method and device, equipment and a storage medium. The image uncertainty prediction method comprises the steps: obtaining a training image set,marking each image in the training image set, and obtaining a marked image set; initializing a preset deep learning model, wherein the preset deep learning model comprises a distributed sampling network and an image segmentation network; training the preset deep learning model based on the training image set and the annotation image set to obtain a first prediction model; extracting a second prediction model from the first prediction model; and obtaining a target image, and performing image uncertainty prediction on the target image through the second prediction model. The invention providesan uncertainty prediction method, capable of obtaining diversified Monte Carlo samples through one-time forward calculation, and enabling uncertainty estimation to be more accurate while reducing thecalculated amount.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to an image uncertainty prediction method, device, equipment and storage medium. Background technique [0002] In current deep learning neural networks such as classification, detection, and segmentation, it is necessary to use an activation function for the last fully connected layer to perform probability mapping operations. The activation function chooses softmax or sigmoid as the ideal probability mapping function without exception. The output result is mapped back to [0, 1] as the output probability value of the algorithm. For example, when the input value is in a certain range, the output value will fluctuate extremely unstable. After the activation function, the uncertainty of the floating disappears, and a stable result is replaced. It can be seen that when the activation function is used as a follow-up process, this uncertainty cannot be well descri...

AI Technical Summary

Problems solved by technology

It can be seen that when the activation function is used as subsequent processing, this uncertainty cannot be well described in the interval of high confidence

[0003] Therefore, in practical applications, the "measure" given by the algorithm is not accurate. For example, in the field of clinical applications, the model judges a disease based on the input image and gives a higher probability, but directly outputs the judgment Probability is unreliable as confidence, so a new confidence metric is needed to quantify the results

Images