Hyperspectral target detection method based on L1 regular constraint depth multi-example learning

Abstract

The invention discloses a hyperspectral target detection method based on L1 regular constraint depth multi-example learning, and mainly solves the problem of poor hyperspectral target detection effectin a complex scene in the prior art. The method comprises the following steps: 1) dividing an input image to obtain a training set and a test set; 2) constructing a deep multi-instance learning network W based on L1 regular constraint; 3) performing iterative training on the network W by using the training set, obtaining a preliminarily trained network Wi when the constructed loss function reaches a minimum value, and inputting the training set into the preliminarily trained network Wi again to perform iterative training to a set maximum number of iterations to obtain a finally trained network W '; and 4) inputting each example point of the test set into the finally trained network W' for detection to obtain a detection result. According to the method, the detection result of the hyperspectral target which is inaccurately marked is improved, the over-fitting phenomenon is reduced, and the method can be used for fine classification of explosives and crops.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a hyperspectral target detection method, which can be used for fingerprint detection in agriculture, explosive detection and crime scenes. Background technique [0002] Hyperspectral image HSI has a rapid growth in the application of remote sensing, and data acquisition is performed by a variety of hyperspectral sensors. Hyperspectral data contains hundreds of continuous spectral bands, which have better analysis capabilities than general images. It acquires spectral curves in many adjacent continuous and very narrow spectral bands, making it possible to construct The essentially continuous radiation spectrum, the sensor simultaneously captures the spatial information of the ground material, forming a unique three-dimensional cube data form. In hyperspectral data, each ground material can be suitably represented by a single spectral curve, called a "spectral...

AI Technical Summary

Problems solved by technology

The disadvantage of this method is that when the training data is unbalanced, it is difficult for the standard SVM classifier to obtain a good classification effect, and when the amount of data is large, the solution process is long

The disadvantage of this method is that when the amount of data is large, the memory space required for the feature map is huge, and it takes a long time to train the SVM classifier.

Images