Small sample underwater target identification method based on deep forest

A technology of underwater target and recognition method, which is applied in the direction of neural learning method, vibration measurement in fluid, measurement vibration, etc., can solve the problem of unsatisfactory target recognition in water, and achieve the effect of improving the recognition accuracy

Pending Publication Date: 2021-08-13

NORTHWESTERN POLYTECHNICAL UNIV

7 Cites 0 Cited by

AI-Extracted Technical Summary

Problems solved by technology

Aiming at the problem that the recognition effect of the traditional machine learning method and the deep learning method based on the neural network are not ideal for underwater targets in the case of small samples, the present invention adopts the Mel frequency cepstral coefficient MFCC feature of the radiation noise...

Abstract

The invention provides a small sample underwater target identification method based on a deep forest. The method comprises the steps that the Mel-frequency cepstral coefficient MFCC features of an underwater target radiation noise signal is taken as the input data of a deep forest model; the deep forest model adopts a cascade structure of a forest to realize layer-by-layer processing of input data so as to perform characterization learning and perform prediction according to output of the last layer of the cascade forest, meanwhile, the number of layers of the cascade forest in the deep forest model is adaptively adjusted according to the data, the model complexity matches the data, and the method is suitable for small sample scenes. The method based on the deep forest provided by the invention can effectively improve the recognition accuracy of the underwater target under the small sample condition.

Application Domain

Vibration measurement in fluidNeural learning methods

Technology Topic

PhysicsAlgorithm +4

Image

Examples

Experimental program(1)

Example Embodiment

[0035] Example

[0036] combined with figure 1 and attached figure 2 , the present invention is based on the deep forest-based small sample water target identification method, comprising the following steps:

[0037] Step 1, place the hydrophone in the ocean, collect and record the radiation noise signals of 6 types of underwater targets respectively, and segment the collected signals in units of 5s to obtain the radiation noise signal sample sets of the 6 types of underwater targets;

[0038] Step 2: Extract the Mel frequency cepstral coefficient MFCC feature of each radiation noise signal sample after segmentation, and obtain the feature vector corresponding to each radiation noise signal sample, and the dimension of the feature vector is selected as 20; when dividing the frame, the frame length is selected. 2048, select 512 for frame shift; select Hanning window for windowing; select 2048 for the number of FFT points; select 128 for the number of filters in the Mel filter bank; select 16000 for the sampling rate of the signal;

[0039] Step 3, label the feature vectors obtained in step 2 according to the target category, and select 0, 1, 2, 3, 4, and 5 for the labels to obtain a data set;

[0040] Step 4: After randomly shuffling the data set obtained in Step 3, divide it by 7:3 to form a training set and a test set;

[0041] Step 5: Construct a deep forest model that only includes cascaded forests. The specific steps are as follows:

[0042] Step 5.1, build 2 completely random tree forests and 2 random forests as the first layer of the cascade forest; each completely random tree forest contains 100 completely random trees; each random forest contains 100 decision trees;

[0043] Step 5.2, build the second layer of the cascade forest, the structure is the same as that of the first layer of the cascade forest described in step 5.1; the output of the first layer of the cascade forest described in step 5.1 is spliced on the original input MFCC feature vector as The input of the second layer of the cascade forest; and so on, a total of 20 layers are constructed;

[0044] Step 6, train the deep forest model described in step 5 on the training set described in step 4; after each layer of cascade forest is added, the performance of the model is verified on the training set, and the training is stopped when the improvement of the verification performance does not exceed 0.001% , to obtain the optimal number of cascaded forest layers, thereby obtaining the optimal deep forest model;

[0045]Step 7, test the optimal deep forest model trained in step 6 on the test set described in step 4, and obtain the accuracy rate of target recognition in water for the test set.

PUM

Description & Claims & Application Information

We can also present the details of the Description, Claims and Application information to help users get a comprehensive understanding of the technical details of the patent, such as background art, summary of invention, brief description of drawings, description of embodiments, and other original content. On the other hand, users can also determine the specific scope of protection of the technology through the list of claims; as well as understand the changes in the life cycle of the technology with the presentation of the patent timeline. Login to view more.

Leakage sound emission signal identification method based on multi-scale morphological decomposition energy spectrum entropy and support vector machine

InactiveCN105469049AStrong resolutionImprove recognition accuracy

Owner:KUNMING UNIV OF SCI & TECH

Classification and recommendation of technical efficacy words

Improve recognition accuracy

Small sample underwater target identification method based on deep forest

AI-Extracted Technical Summary

Problems solved by technology

Abstract

Image

Examples

Example Embodiment

PUM

Description & Claims & Application Information

Similar technology patents

Video character recognition method and device, storage medium and electronic equipment

Mobile robot, working surface recognizing method and control method

Method for detecting insulator string breakage defect based on sparse representation

Electric power load identification method, device and apparatus, and readable storage medium

Leakage sound emission signal identification method based on multi-scale morphological decomposition energy spectrum entropy and support vector machine

Classification and recommendation of technical efficacy words

Barrier identification method and system of laser radar

Semi-supervised-transfer-learning character recognition method and system based on convolutional neural network

Identification method of fake-licensed cars

Non-intrusive resident load identification method based on S_Kohonen

Vehicle fault detection system and method based on vehicle-mounted CAN network