A log detection method based on weight sharing and upper and lower feature fusion

Abstract

The invention proposes a log detection method based on weight sharing and adjacent upper and lower feature fusion models, which belongs to the field of machine vision and deep learning. This method first constructs a pooled feature pyramid to realize weight sharing features, and secondly simplifies the top-down feature fusion to two adjacent layers for feature fusion, which reduces the complexity of model fusion and the transmission of redundant information; Both improvement methods lead to a significant improvement in the accuracy of the detector, especially for the self-made log dataset. The present invention fully considers the characteristics of dense small targets and lack of data, improves the robustness and training speed of the model through feature fusion and weight sharing, and has a more reliable source of analysis data in practical applications.

Description

A log detection method based on weight sharing and upper and lower feature fusion technical field The invention belongs to the field of machine vision and deep learning, relates to deep learning target detection and recognition technology, especially It involves a log detection method based on weight sharing and upper and lower feature fusion, and proposes to use the maximum pooling operation to construct pooling Feature pyramid, which enables features to transmit information flow from bottom to top, and realizes the log detection method of feature sharing. Background technique [0002] In the traditional SSD algorithm, the feature maps of different layers are independently used as the input of the classification network, so there may be When the same object is detected by frames of different sizes at the same time, the detection effect on small scales is relatively poor. Currently through The top-down deconvolution performs feature fusion, so that the feature informa...

AI Technical Summary

Problems solved by technology

[0002] In the traditional SSD algorithm, the feature maps of different layers are independently used as the input of the classification network, so the same object may be detected by frames of different sizes at the same time, and the detection effect on small scales is relatively poor.

Images