CN105469097A - Transformer station feature extraction method based on nerve network - Google Patents

Abstract

The invention discloses a transformer station feature extraction method based on the nerve network. The method comprises steps that, a first step, image pre-processing and feature segmentation are carried out; a second step, calculation for a contour shape parameter F and attribute marking are carried out; a third step, N transformer station field images are acquired randomly, each image is processed according to the first step and the second step to acquire a data matrix of the contour shape parameter F classified according to contour attributes; a fourth step, BP network initialization is carried out; a fifth step, BP network training is carried out; and a sixth step, automatic identification on to-be-detected image attributes is accomplished. Through the transformer station feature extraction method based on the nerve network, accuracy and correctness of analysis on the transformer station video images are improved, and the utilization value of the transformer station image data is improved.

Description

A kind of transformer station's feature extracting method based on neural network

Technical field

The present invention relates to a kind of transformer station's feature extracting method based on neural network, belong to technical field of image processing.

Background technology

At present, a large amount of indoor and outdoor equipment that video monitoring system exists transformer station, carry out intuitive monitoring in scheduling or center of the centralized monitor to it.Video monitoring system improves transformer substation security ability.Numerous objects such as transformer, isolating switch, bus, electric wire, insulator, construction thing, steelframe are usually comprised in substation field video image, how accurately and effectively numerous transformer stations object to be carried out distinguishing, extracting from video image, realize real unmanned, the key that intelligence is on duty.

Summary of the invention

Object: in order to overcome the deficiencies in the prior art, the invention provides a kind of transformer station's feature extracting method based on neural network.

Technical scheme: for solving the problems of the technologies described above, the technical solution used in the present invention is:

Based on transformer station's feature extracting method of neural network, comprise step as follows:

Step one: the pre-service of image and Image Segmentation Methods Based on Features, carries out image gray processing, image background is removed, image irradiation compensation deals by substation field image; Split in the substation field image after process respectively by profile about power equipment and personnel's each several part;

Step 2: calculate contour shape parameter F and attribute flags, each several part profile is calculated by formula (1); And manually mark by profile actual attribute;

$F=\frac{\left|\right|B|{|}^2}{4\mathrm{\π}A}---\left(1\right)$

B is profile length, and A is contour area;

Step 3: get N at random and open substation field image, often opens the process that image carries out step one, step 2, obtains the data matrix of the contour shape parameter F by profile attributes classification;

Step 4: BP netinit, by little random number to the weight w of every one deck and deviation b initialization, to ensure that network is unsaturated by large weighting input;

Step 5: BP network training, the data matrix obtained in step 3 input BP network is trained, obtain BP network model parameter: training step is 100, the intervening step of display training result is 25, training objective error is 0, the training permission time is Inf, and in training, minimum permission Grad is LE-6;

Step 6: the contour shape parameter F of testing image is inputted BP netinit, completes testing image attribute and automatically identify.

Beneficial effect: a kind of transformer station's feature extracting method based on neural network provided by the invention, improves accuracy and the correctness of transformer substation video graphical analysis, improve the value that view data is looked by transformer station.

Accompanying drawing explanation

Fig. 1 is structural representation of the present invention.

Embodiment

Below in conjunction with accompanying drawing, the present invention is further described.

As shown in Figure 1, a kind of transformer station's feature extracting method based on neural network, comprises step as follows:

Step one: the pre-service of image and Image Segmentation Methods Based on Features, carries out image gray processing, image background is removed, image irradiation compensation deals by substation field image; Split in the substation field image after process respectively by profile about power equipment and personnel's each several part;

Step 2: calculate contour shape parameter F and attribute flags, each several part profile is calculated by formula (1); And manually mark by profile actual attribute;

$F=\frac{\left|\right|B|{|}^2}{4\mathrm{\π}A}---\left(1\right)$

B is profile length, and A is contour area;

Step 3: get N at random and open substation field image, often opens the process that image carries out step one, step 2, obtains the data matrix of the contour shape parameter F by profile attributes classification;

Step 4: BP netinit, by little random number to the weight w of every one deck and deviation b initialization, to ensure that network is unsaturated by large weighting input;

Step 5: BP network training, the data matrix obtained in step 3 input BP network is trained, obtain BP network model parameter: training step is 100, the intervening step of display training result is 25, training objective error is 0, the training permission time is Inf, and in training, minimum permission Grad is LE-6;

Step 6: the contour shape parameter F of testing image is inputted BP netinit, completes testing image attribute and automatically identify.

The above is only the preferred embodiment of the present invention; be noted that for those skilled in the art; under the premise without departing from the principles of the invention, can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (1)

1., based on transformer station's feature extracting method of neural network, it is characterized in that: comprise step as follows:

Step one: the pre-service of image and Image Segmentation Methods Based on Features, carries out image gray processing, image background is removed, image irradiation compensation deals by substation field image; Split in the substation field image after process respectively by profile about power equipment and personnel's each several part;

Step 2: calculate contour shape parameter F and attribute flags, each several part profile is calculated by formula (1); And manually mark by profile actual attribute;

$F=\frac{\left|\right|B|{|}^2}{4\mathrm{\π}A}---\left(1\right)$

B is profile length, and A is contour area;

Step 3: get N at random and open substation field image, often opens the process that image carries out step one, step 2, obtains the data matrix of the contour shape parameter F by profile attributes classification;

Step 4: BP netinit, by little random number to the weight w of every one deck and deviation b initialization, to ensure that network is unsaturated by large weighting input;

Step 5: BP network training, the data matrix obtained in step 3 input BP network is trained, obtain BP network model parameter: training step is 100, the intervening step of display training result is 25, training objective error is 0, the training permission time is Inf, and in training, minimum permission Grad is LE-6;

Step 6: the contour shape parameter F of testing image is inputted BP netinit, completes testing image attribute and automatically identify.