Skin lesion segmentation and feature extraction method based on deep residual pyramid

Abstract

The invention relates to a skin lesion segmentation and feature extraction method based on a deep residual pyramid. The method comprises the following steps of 1, constructing a skin disease intelligent extraction model based on the deep residual pyramid; 2, constructing a deep residual pyramid multi-scale coding network, training the extracted model through the residual pyramid multi-scale coding network and a training server, extracting multi-scale features, and outputting segmentation and extraction results; and 3, designing a loss gradient supervision mechanism based on focus loss and a gradient coordination mechanism, balancing weights of outliers, easy examples and hard examples, and making the model properly pay attention to the hard examples. According to the method, a deep residual network and a feature pyramid are combined, the deep residual pyramid multi-scale coding network is constructed, a bottleneck layer is divided into the multi-scale coding network, network segmentation and extraction result output is realized by extracting the multi-scale features, and the segmentation capability and feature extraction capability of a neural network are improved.

Description

technical field [0001] The invention relates to the field of artificial intelligence detection of skin diseases, in particular to a skin lesion segmentation and feature extraction method based on a deep residual pyramid. Background technique [0002] Dermatosis is one of ubiquitous diseases, and distribution age is wide. Malignant skin lesions, such as melanoma, are one of the most malignant cancers in the world due to their rapid progression and high fatality rate. Studies have shown that if melanoma can be screened and found at an early stage, this type of tumor can be completely cured. Therefore, efficient and timely screening of malignant skin diseases such as melanoma has important medical value. The identification and classification of melanoma is very difficult due to the wide variety of skin diseases, high inter-class similarity and large intra-class difference in appearance of lesions. In recent years, the diagnosis and death cases of melanoma have continued to i...

AI Technical Summary

Problems solved by technology

However, such ensemble models are resource-intensive and expensive to train

[0009] Aiming at the problems of poor network extraction ability and low accuracy, some scholars adopt the method of increasing network depth

Although this type of method improves the performance of the model by increasing the depth of the network, as the depth of the network increases, the accuracy of the network appears saturated or even declines, the network begins to degenerate, and the extracted features are easily lost.

Images