Multispectral Target Recognition Based On Pyramid Layered Model

With the high-speed development of artificial intelligence,general target detection for a class of specific targets plays a vital part in the machine vision.Those traditional classifier models,however,require a lot of training samples,which may lead to slow learning process.What's more,phenomena of over-fit training parameters may easily occur.Therefore,new ways of target detection without training have been developed in recent years.This paper focuses on the multispectral image detection,and puts forward two robust target detection model.(1)Pyramid hierarchical structure model based on the LARK.In this model,we construct the 3D-LARK feature descriptors by fully using the spatial structure and the abundant spectral information,in order to solve the low recognition accuracy caused by the loss of spectral information in gray images.Meanwhile,for the complex calculation and low accuracy,we propose the pyramid hierarchical model.For the changeable targets in images,a composite template set is established to ensure all targets can be detected,which is multi-scale and multi-angle.Experiments show that,compared with the training-free algorithm for gray images,the proposed model uses the space and spectrum information effectively,and reduces the interference of the local similar structure in the background noise,as well as the error detection rate.(2)Target detection model based on the local structure and neighbor constraint.Relied on the above model,the neighbor structure characteristic is further introduced to optimize the integrity of the detection results.What's more,with the neighbor structure characteristic and 3D-LARK feature descriptors,the alternating detections is conducted in the pyramid model to improve the detection accuracy,so as the efficiency.Experiments are taken in the standard library with cars and pedestrian,and compared with other training-free algorithm.It can be proved that,the proposed model is effective in color image target detection,and can reduce the interference of background noise and solve the problem of overly split feature.