Visual researchers at the University of York have found that the human brain does not prioritize interesting areas of images.

Visual researchers at the University of York in Canada recently tested a variety of visual image processing algorithms and found that the human brain does not prioritize interesting areas of the image, challenging the 61-year-old classic theory, according to a new report by the Physicist Network.
this achievement is of great theoretical significance in understanding human vision, visual processing and diagnosis of visual pathology, and will also play a role in the establishment and improvement of applied models such as autonomous driving.
related paper was published in the recent journal PLOS ONE.
team of John Sotts, a professor in the Department of Electrical Engineering and Computer Science at the University of York's Lassond School of Engineering, found that the human brain does not prioritize interesting areas of the image.
this runs counter to the widely influenced "early choice theory" by psychologist Professor Donald Broadbent.
the theory, the brain follows and processes the relevant information in turn, according to how interesting the image is.
hundreds of algorithms have been established based on the interesting extent of this image area, but this theory has not been questioned. "Our study tested hundreds of algorithms that were more advanced and asked the question——— 'What are the performance of these algorithms and the performance of humans?'" said
Sotts. The researchers designed some repeated experiments.
subjects in one of the experiments had to tell whether animals were drawn in a particularly processed image in a short period of time, and the subjects could only give answers by using images of different small areas of the picture.
these images did not show a difference in the degree of interestingness, thus eliminating the direct impact of the parameter of remarkableness.
subjects were able to quickly and correctly identify the results of the image content, contrary to theoretical "settings".
results show that the ability of humans to determine what an image depicts has nothing to do with the remarkable degree of image fun, and that algorithms based on the rules of remarkablity are much less able to recognize than human performance.
other experiments have also shown that while remarkableness mainly determines which scene the human eye sees first, it is the eye movement that determines the order in which the human brain processes.
in other words, the remarkable cannot determine the order in which the brain processes the vision.
Microbank AI team won the National Natural Science Foundation satellite remote sensing image intelligent analysis competition double track champion recently, the National Natural Science Foundation of China Information Science Department, "space information network basic theory and key technology" major research project guidance expert group sponsored by the "remote sensing image sparse characterization and intelligent analysis competition" came to an end, micro-bank AI team in remote sensing image scene classification track pressure competitors, won the track champion.
at the same time, the AI team of Microbank and the Academy of Artificial Intelligence of Xi'an University of Electronic Science and Technology also won the title in the target tracking track.
it is understood that this is the third competition organized by the National Natural Science Foundation of China, set up remote sensing image scene classification, remote sensing image target detection, remote sensing image semantic segmentation, remote sensing image change detection and remote sensing video target tracking five competition topics, attracted 11 countries, 115 cities 2191 from major universities, research institutes and technology companies related fields of research team participation.
remote sensing image scene classification track as one of the five tracks of the whole competition, with remote sensing images containing typical scenes as the processing object, the team used the data provided by the organizers to classify the designated remote sensing images, and the organizers evaluated the results of remote sensing image scene classification according to the scoring criteria.
competition, a remote sensing image dataset containing 45 categories of scenes was provided, with a sample size of more than 200,000.
the sample style of the same scene is very different, and the sample number is not balanced between categories, which is very high on the algorithm. In addition,
, because remote sensing images generally contain a wide geographical scope, geographical situation is complex and diverse, mountains and lakes are connected, cities and farmland, different types of scenes, often concentrated in the same photo, how to define the scene becomes difficult.
especially many scenarios, such as deserts and oil fields, urban intersections and bridges, pose higher challenges to technology and algorithms.
, the AI team of Microbank designed a series of solutions such as data enhancement scheme, expert model and multi-model fusion, which greatly improved the accuracy and accuracy of recognition.
scene classification technology is one of the basic techniques of remote sensing image analysis, which is widely used in intelligent cities and smart agricultural scenes such as urban development analysis, agricultural surface classification and functional area delimitation.
based on the core AI technology such as remote sensing image analysis technology, the AI team of Microbank set up an intelligent asset management platform, through the analysis of remote sensing images, Internet of Things data, public opinion data and other "alternative data" analysis, mining more real-time, intelligent and comprehensive information, so as to more sensitivemonitor monitoring macroeconomic trends, quantitative fundamental analysis, tracking commodities, agricultural insurance damage and ESG rating of enterprises.
, for example, by satellite remote sensing images, to detect the targets of vehicles, ships, aircraft, crude oil tanks and so on in the region, and then to estimate the business conditions of enterprises, commodity production, etc., and by identifying the geomorphological characteristics of different crops, the area of crop cultivation, growth and so on can be monitored and estimated.
more financial scenarios and solutions based on remote sensing images will be showcased in the upcoming "Micro-AI Intelligent Management Platform".
micro-bank AI team is committed to building intelligent management technology, through the industry-leading artificial intelligence technology mining huge amounts of alternative data, is expected to solve the asset management business pain points and difficulties.
artificial intelligence is moving towards the critical point of technological inclusion, the era of pan-artificial intelligence is coming.
the future, artificial intelligence technology will affect the daily production life style of users in a more in-depth manner, reshape the production structure, and promote the further development of industrial intelligence.
Microbank AI team will continue to use AI technology to benefit the public, the real AI technology to achieve a deeper landing, promote economic and efficient development, achieve inclusive AI, enabling life.
Source: Science Daily.