A Weakly Supervised Method for Mud Detection in Ores Based on Deep Active Learning

Algorithm 1

The YOLO-v3-based AL framework.

	Inputs:
	(labeled training image sample set)
	(unlabeled image set)
	(the pretrained YOLO-v3 model)
	(the sample increment)
	Outputs:
	(fine-tuned by AL method)
(1)	Train the YOLO-v3 model with the labeled training image sample set and update and YOLO-v3 detector.
(2)	Detect objects for each image in and calculate confidence for each object.
(3)	Sort the mud objects in ascending order according to and take the first mud objects contained in images which is composed of sample set .
(4)	For each image in , the expert checks the object label or box bound and makes appropriate modification. The images with verified labels form sample set .
(5)	Add verified samples to the current training set and remove them from .
(6)	Continue step 1 to step 5 till the set is null or reaches the specified number of iterations.