Method for image feature deep learning and perceptibility quantification through computer

Abstract

The invention discloses a method for image feature deep learning and perceptibility quantification through a computer, and the method comprises the following steps: A, collecting street images; B, building a scoring model, carrying out the crowdsourcing scoring of the perceptibility of a part of collected street images, carrying out the sorting according to the value of perceptibility, and collecting the scoring data; C, carrying out the training of a convolution neural network classifier through the scoring data collected at step B, and carrying out the prediction and scoring of the perceptibility of the remaining images through the trained convolution neural network classifier; D, predicting the street visualization perceptibility on a map, and giving the score estimation of different perceptibility values of streets. The method is simple and convenient, can greatly shorten the training time in executing a task, guarantees the stable and reliable overall performances, can achieve the effective recognition of the image perceptibility and the large-scale quantitative perceptibility analysis, greatly enlarges the application range, and is good in application stability, applicability and practicality.

Description

technical field [0001] The invention relates to a method for computer deep learning of image features and quantification of perception. Background technique [0002] Image recognition and object detection are important research issues in the field of computer vision, and have broad application prospects in many aspects such as face recognition, security monitoring, and dynamic tracking. Image recognition refers to the technology of using computers to process, analyze and understand images to identify targets and objects in various patterns. Object detection refers to detecting and identifying a specific target in any frame or continuous frame images, and returning the position and size information of the target, such as outputting a bounding box surrounding the target. [0003] At present, convolutional neural network (CNN) has been widely used in image classification, object detection and other problems, without any human intervention, to extract information from images an...

AI Technical Summary

Problems solved by technology

[0004] Therefore, the existing technology will have the following defects: (1) It is necessary to train the model on a huge data set for a specific task, resulting in long pre-training time and poor overall performance when performing the task

(2) The computational complexity of CNN is high, which limits its use in application scenarios with high real-time requirements; (3) It is impossible to extract the depth and level features of images, so it is impossible to recognize the perception of pictures and perform large-scale Quantitative perception analysis of

Therefore, the applicability and practicability are limited, and it is difficult to meet the needs of the market

Images