Deep learning plays an important role in artificial intelligence (AI), which yet faces challenges in object representation. In particular, with object sizes decreasing, recognition and detection become more difficult and inaccurate.

A research team led by WANG Hongqiang from the Hefei Institutes of Physical Science (HFIPS) of the Chinese Academy of Sciences (CAS) has recently proposed a strategy to stimulate the homocentric opponent phenomenon of human eye in the field of vision.

According to their recently published paper on Neurocomputing, effective field-of-view theory is used to enhance the AI representation learning model and has been proved in the research to significantly improve target detection performance.  

At the outset, the researchers noticed that there is always an effective receptive field (eRF) when eyes are paying attention to something.  

Mimicking this, they designed an eRF module for deep neural networks, which preserve training samples within the range of an object's effective reception domain. At the same time, the size of effective receiving domain is adjusted adaptively for different backbone layers for sample balance.  

Compared with current products on the market, this model enhances object representation of deep networks, with less training time and higher accuracy. Through comparative experiment, the average accuracy of the proposed model is 32.5%, which is 10% higher than that of the traditional model. The key is that the performance of this model is better than other models on small target detection.  

This work was supported by the National Natural Science Foundation of China and the Key Research and Development Program of Anhui Province.

Numerical simulation of the change of the effective field of view of the network. (Image by WANG Hongqiang)