CN-121505373-B - Model training method, device, visual assessment method, aircraft and medium

CN121505373BCN 121505373 BCN121505373 BCN 121505373BCN-121505373-B

Abstract

The application provides a model training method, a device, a visual assessment method, an aircraft and a medium, which relate to the technical field of image processing, the method extracts global contrast, edge definition, texture characteristics and dust concentration associated with an sand blind environment through a sample visual image, and obtaining corresponding sample sand blind visual characteristics, obtaining a target training sample set based on the sample sand blind visual characteristics and corresponding label information, and training the initial visual assessment model based on the target training sample set to obtain a target visual assessment model. The initial visual assessment model can learn enough information related to the sand blind environment through the sand blind visual characteristics, and learn the relevance between the sand blind visual characteristics and subjective assessment of a pilot, so that the target visual assessment model can accurately output visual assessment results conforming to the actual experience of the pilot, and the scientificity and rationality of visual assessment in the sand blind environment can be improved.

Inventors

YANG NINGMENG
ZHANG WEIGUO
WANG LIANGQUAN
SHI ZHEYU
MA SHUAI

Assignees

中国空气动力研究与发展中心低速空气动力研究所

Dates

Publication Date: 20260512
Application Date: 20260114

Claims (5)

1. A method of model training, comprising: The method comprises the steps of obtaining an initial training sample set and an initial visual assessment model, wherein the initial training sample set comprises a plurality of training sample pairs, each training sample pair comprises a sample visual image and label information, and the label information comprises a visual assessment result corresponding to the sample visual image which is marked manually; Extracting features of each sample visual image in the initial training sample set to obtain sample sand blind visual features; Obtaining a target training sample set based on the sample sand blind visual characteristics and the label information corresponding to each training sample; training the initial visual assessment model based on the target training sample set to obtain a trained target visual assessment model, wherein the target visual assessment model is used for visually assessing the blindness visual characteristics corresponding to the blindness environmental image; feature extraction is carried out on each training sample in the initial training sample set to obtain sample sand blind visual features, and the method comprises the following steps: extracting features of the sample visual image to obtain global contrast, edge definition, texture features and sand and dust concentration; Respectively normalizing the global contrast, the edge definition, the texture feature and the sand concentration to obtain normalized features, wherein the normalized features comprise global contrast normalization values Edge definition normalization value Normalized values of texture features Normalized sand and dust concentration value ; Determining weight parameters based on the normalized features, wherein the weight parameters comprise first weights corresponding to global contrast normalization values Second weight corresponding to edge definition normalization value Third weight corresponding to texture feature normalization value Fourth weight corresponding to normalized sand and dust concentration value ; The weight parameter determination formula is: ; ; wherein C represents global contrast, E represents edge definition, T represents texture features, and H represents sand and dust concentration; As a degradation factor, the degradation factor is, Is a super-parameter that is greater than 0, The greater the value, the more concentrated the weights are assigned to the one or two features that are most degraded; is a natural exponential function; and determining the sample sand blind visual characteristics based on the normalized characteristics and the weight parameters.
2. A visual assessment method, comprising: when the current flight environment is an sand blind environment, acquiring a current visual image; extracting the characteristics of the current visual image to obtain current sand blind visual characteristics; And obtaining a current visual assessment result based on the current sand blind visual characteristics and a target visual assessment model, wherein the target visual assessment model is trained based on the model training method of claim 1.
3. Model training apparatus for performing the method according to claim 1 or 2, comprising: The system comprises a data acquisition module, a visual evaluation module and a visual evaluation module, wherein the data acquisition module is used for acquiring an initial training sample set and an initial visual evaluation model, the initial training sample set comprises a plurality of training sample pairs, each training sample pair comprises a sample visual image and label information, and the label information comprises a visual evaluation result corresponding to the sample visual image which is marked manually; The device comprises a feature extraction module, a feature extraction module and a test module, wherein the feature extraction module is used for carrying out feature extraction on each sample visual image in an initial training sample set to obtain sample sand-blind visual features, and the sand-blind visual features comprise global contrast, edge definition, texture features and sand concentration; the sample acquisition module is used for acquiring a target training sample set based on the sample sand blind visual characteristics and the label information corresponding to each training sample; The model training module is used for training the initial visual assessment model based on the target training sample set to obtain a trained target visual assessment model, wherein the target visual assessment model is used for performing visual assessment on the sand blind visual characteristics corresponding to the sand blind environment image; the feature extraction module comprises: The feature extraction unit is used for extracting features of the sample visual image to obtain global contrast, edge definition, texture features and dust concentration; And the characteristic determining unit is used for determining the sample sand blind visual characteristic based on the global contrast, the edge definition, the texture characteristic and the sand dust concentration.
4. An aircraft comprising a memory and a processor for executing program instructions stored in the memory to implement the visual assessment method of claim 2.
5. A computer readable storage medium, having stored thereon a computer program which, when executed by a processor of a computer, causes the computer to perform the method of claim 1 or 2.

Description

Model training method, device, visual assessment method, aircraft and medium Technical Field The application relates to the technical field of image processing, in particular to a model training method, a model training device, a model visual assessment method, an aircraft and a model visual assessment medium. Background When the helicopter flies near the ground in loose ground surface environments such as deserts, gobi and the like, a large amount of sand and dust can be rolled up by the downward washing flow of the rotor wing, so that a sand and dust cloud which shields the sight is formed, namely, the phenomenon of 'sand blindness'. The sand blindness causes the pilot to lose visual reference, so that flight accidents are extremely easy to be caused, and flight safety is seriously threatened. Currently, studies on the phenomenon of sand blindness have focused on two aspects: 1. sand lifting mechanisms and suppression techniques such as simulating sand motion trajectories by Computational Fluid Dynamics (CFD) and Discrete Element Method (DEM) coupling, or suppressing sand lifting by optimizing rotor layout parameters (e.g., tip shape, dihedral angle). 2. Simple visibility assessment the prior art generally crudely estimates visibility by physically calculating the concentration of sand or counting the number of sand particles in a unit field of view. The above studies focus on the physical properties of the "dust" itself, but ignore the quantitative analysis of the visual perception of the pilot. The existing evaluation method cannot fully consider the degradation rule of the human eye vision system in a sand dust environment. Therefore, the actual influence degree of the sand blind environment on the subjective visual feeling and the situational awareness of the pilot cannot be accurately estimated, and the sand blind environment is difficult to be effectively applied to high-fidelity flight simulation, pilot training estimation and auxiliary landing system development. Disclosure of Invention The application provides a model training method, a device, a visual assessment method, an aircraft and a medium, global contrast, edge definition, texture characteristics and sand concentration associated with an sand blind environment are extracted through sample visual images, and obtaining corresponding sample sand blind visual characteristics, obtaining a target training sample set based on the sample sand blind visual characteristics and corresponding label information, and training the initial visual assessment model based on the target training sample set to obtain a target visual assessment model. The initial visual assessment model can learn enough information related to the sand blind environment through the sand blind visual characteristics, and learn the relevance between the sand blind visual characteristics and subjective assessment of a pilot, so that the target visual assessment model can accurately output visual assessment results conforming to the actual experience of the pilot, and the scientificity and rationality of visual assessment in the sand blind environment can be improved. The application provides a model training method, which comprises the following steps: The method comprises the steps of obtaining an initial training sample set and an initial visual assessment model, wherein the initial training sample set comprises a plurality of training sample pairs, each training sample pair comprises a sample visual image and label information, and the label information comprises a visual assessment result corresponding to the sample visual image which is marked manually; Extracting features of each sample visual image in the initial training sample set to obtain sample sand-blind visual features, wherein the sand-blind visual features comprise global contrast, edge definition, texture features and sand concentration; Obtaining a target training sample set based on the sample sand blind visual characteristics and the label information corresponding to each training sample; Training the initial visual assessment model based on the target training sample set to obtain a trained target visual assessment model, wherein the target visual assessment model is used for visually assessing the blindness visual features corresponding to the blindness environment image. Optionally, feature extraction is performed on each training sample in the initial training sample set to obtain a sample sand blind visual feature, including: extracting features of the sample visual image to obtain global contrast, edge definition, texture features and sand and dust concentration; sample sand blind visual features are determined based on global contrast, edge definition, texture features, and dust concentration. Optionally, determining the sample sand blind visual features based on the global contrast, edge sharpness, texture features, and sand concentration includes: And splicing the global contrast, edge definition, tex