Instrument recognition method based on improved U
View Patent ↗The present invention discloses an instrument recognition method based on an improved U 2 network. The method includes: replacing common convolution of each layer with grouped convolution Grouped Conv on a basis of an RSU, and segmenting a dial plate and a pointer by using the network; performing noise reduction processing on the scale value array by using a mean filter; and determining a position of a scale value corresponding to the pointer by using a peak value, and outputting a reading according to the scale value and preset data. A pointer-type instrument is segmented by using the improved U 2 network, and automatic reading of the obtained dial plate is implemented by using a conventional computer vision method. Therefore, compared with manual reading, the method has advantages of high precision, high reliability, fast reading, low costs, and the like, and working efficiency can be greatly improved.
1 . An instrument recognition method based on an improved U 2 network, comprising the following steps:
obtaining a to-be-recognized instrument image and preset data of an instrument;
inputting the to-be-recognized instrument image into a trained image segmentation model based on the improved U 2 network, and obtaining, based on an output of the image segmentation model, a masked picture comprising a dial plate scale and a pointer image;
unfolding the circular dial plate scale in the masked picture into a two-dimensional matrix image, and then converting the two-dimensional matrix image into a one-dimensional scale array;
performing noise reduction processing on the scale array by using a mean filter, to obtain the number Point_num of dial plate scale lines passed by a pointer and the total number Nums of dial plate scale lines; and
obtaining a final instrument reading through calculation based on the number Point_num of dial plate scale lines passed by the pointer and the total number Nums of dial plate scale lines, as an instrument recognition result;
wherein the unfolding the circular dial plate scale in the masked picture into a two-dimensional matrix image comprises:
X=cx+r *sin θ, and
Y=cy−r *cos θ,
wherein X and Y are coordinates after unfolding into a rectangle, cx and cy are center coordinates of a circle, r is a radius, and θ is a rotation angle corresponding to a scale line.
2 . The instrument recognition method based on an improved U 2 network according to claim 1 , wherein the obtaining the final instrument reading through calculation based on the number Point_num of dial plate scale lines passed by the pointer and the total number Nums of dial plate scale lines comprises:
Num
=
Point_num
Nums
*
R
,
wherein Num represents the instrument reading, Point_num represents the number of dial plate scale lines passed by the pointer, Nums represents the total number of dial plate scale lines, and R is a range of the instrument.
3 . The instrument recognition method based on an improved U 2 network according to claim 1 , wherein
the image segmentation model based on the improved U 2 network comprises a residual U-shaped block RSU for multi-scale feature extraction and an outer U-shaped structure connected to the RSU, wherein the RSU comprises three parts: an input convolutional layer for performing channel conversion on an input, a U-shaped structure for multi-scale extraction of context information, and a residual connection layer connected to an input layer and an intermediate layer through a residual block; convolution in the residual U-shaped block RSU is grouped convolution Grouped Conv; for grouped convolution Grouped Conv, because an input feature map size is H*W*C 1 /g, C 2 convolution kernels of a size H*W*C 1 /g are required; and because a concatenation operation does not require additional parameters, a parameter quantity is H*W*C 1 /g, which is 1/g of an original quantity.
4 . The instrument recognition method based on an improved U 2 network according to claim 1 , wherein a method for training the image segmentation model based on the improved U 2 network comprises:
obtaining a labeled instrument image data set, and dividing the data set into a training set, a validation set, and a test set, wherein labels comprise the pointer, the dial plate scale, and a background; and
training and testing the image segmentation model based on the improved U 2 network by using the data set, to obtain the trained image segmentation model based on the improved U 2 network.
5 . The instrument recognition method based on an improved U 2 network according to claim 4 , wherein the training the image segmentation model based on the improved U 2 network by using the data set comprises:
after performing data labeling by using labelme, converting the instrument picture in a jpg format into a picture in a single-channel png format, wherein 0 is the background, 1 is the pointer, and 2 is the scale; then inputting the raw picture and the corresponding png picture into the network for training; and then comparing the obtained scale and pointer with the corresponding png picture, and minimizing a difference between the scale and pointer and the png picture by continuously adjusting parameters, to obtain the trained image segmentation model based on the improved U 2 network.
6 . The instrument recognition method based on an improved U 2 network according to claim 1 , wherein the preset data of the instrument comprises scale values, scale division values, and range information of the same instrument.
7 . An instrument recognition apparatus based on an improved U 2 network, comprising a processor and a storage medium, wherein
the storage medium is configured to store instructions; and
the processor is configured to perform operations according to the instructions to perform the steps of the method according to claim 1 .
8 . A non-transitory storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to claim 1 are implemented.