Weakly supervised convolutional neural network image target positioning method

Abstract

The invention discloses a weakly supervised convolutional neural network image target positioning method, which comprises the following steps of: establishing a convolutional neural network classification model with a batch normalization layer, training the convolutional neural network classification model, and storing the trained convolutional neural network classification model; s2, inputting ato-be-positioned image into the convolutional neural network classification model trained in the step S1, and obtaining a feature map output by the deep convolutional layer; performing weighted fusionon the obtained feature map to obtain a saliency map; converting the obtained saliency map into a thermodynamic diagram, and superposing the thermodynamic diagram on the input image to generate a composite image; and storing or visualizing the obtained composite image to obtain a target positioning image. According to the weak supervision convolutional neural network image target positioning method using the batch normalization scaling factor as the corresponding feature map weight, the problems that in the prior art, implementation is complex, category confidence scores, gradients and otherinformation are needed, a convolutional neural network model is not transparent, and the working function is poor are solved.

Description

technical field [0001] The invention relates to the field of image positioning, in particular to a weakly supervised convolutional neural network image target positioning method. Background technique [0002] The convolutional neural network first made a breakthrough in the image classification task, and because of its outstanding feature extraction ability, it has been widely used in various fields, such as image target positioning. When the convolutional neural network is applied to image classification, it only needs to simply encode the image category, while in the target positioning task, it is necessary to manually mark the target position in the image with a bounding box in advance. Therefore, compared with image classification tasks, target localization tasks require stronger supervision and are more challenging. [0003] Relevant scholars have proposed a weakly supervised image target location method based on class activation maps, using the weight parameters of th...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the above-mentioned deficiencies in the prior art, the present invention provides a weakly supervised convolutional neural network image target positioning method, which solves the need for bounding box labeling information and the need for backpropagation calculation and model modification in existing target positioning methods. structure and lead to complex implementation problems, and at the same time solve the problem that the opaque working mechanism of the convolutional neural network model cannot be well explained

Images