Algorithm for predicting pyramid feature map

Abstract

The invention discloses an algorithm for predicting a pyramid feature map. The algorithm comprises the following steps of: selecting and dividing a data set, and carrying out preprocessing operation on pictures input by the data set; modifying a head part of an FCOS-like network architecture and setting the head part as a new network structure; sending the input data into a backbone network to obtain a feature map of the input data, carrying out regression operation on each point of the feature map, and carrying out network training to obtain a network model; applying a pre-trained network model to the test picture, and obtaining a prediction result from a plurality of Heads of the feature pyramid; according to the method, the head network of a prediction network part is improved by adopting a lightweight simple model to obtain the prediction results of the feature maps on different levels, and then the prediction results of the pyramid feature maps on different levels are combined toobtain the final prediction result, so that the accuracy of the method is remarkably improved compared with that of an SOAT method in the current target detection field.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an algorithm for predicting pyramid feature maps. Background technique [0002] In recent years, due to the development and application of convolutional neural network (referred to as CNN), many tasks in the field of computer vision have been greatly developed, among which target detection is an important task in computer vision. Object detection is a popular direction in computer vision and digital image processing. It is widely used in robot navigation, intelligent video surveillance, industrial inspection, aerospace and many other fields. It has important practical significance to reduce the consumption of human capital through computer vision. Therefore, target detection has become a research hotspot in theory and application in recent years. It is an important branch of image processing and computer vision, and it is also the core part of intelligent monitoring syst...

AI Technical Summary

Problems solved by technology

At present, most of this kind of detectors are based on a point of the bounding box to make predictions, and the features obtained in this way may not be representative, so some people later proposed to introduce more and stronger features, but many of these increased operations will be due to Introduces background noise with harmful information

However, these methods do not explicitly extract boundary features. This application believes that the boundary limit point features are more important for the positioning of the bounding box, and select the most effective features at each boundary, and use a concise and efficient method to complete the feature extraction. Therefore, , the existing technology needs to be further improved and perfected

Images