CN110705604B - AI algorithm-based dynamic model detection method - Google Patents
AI algorithm-based dynamic model detection method Download PDFInfo
- Publication number
- CN110705604B CN110705604B CN201910860225.7A CN201910860225A CN110705604B CN 110705604 B CN110705604 B CN 110705604B CN 201910860225 A CN201910860225 A CN 201910860225A CN 110705604 B CN110705604 B CN 110705604B
- Authority
- CN
- China
- Prior art keywords
- input image
- component
- dynamic model
- detection method
- algorithm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention relates to a dynamic model detection method, in particular to a dynamic model detection method based on AI algorithm.A system receives a picture evaluation request, acquires an input image from a video access device, carries out sharpening processing on the input image, loads a training model and a data set sample label, acquires a tensor in the training model, converts the input image into a numpy array format, carries out feature extraction on the input image by using a CNN convolutional layer, generates a suggestion window by using a Regional Proposal Network (RPN), maps the suggestion window to the last layer of convolutional feature map of the CNN convolutional feature map, and enables each region of interest to generate a fixed-size ROI feature map by using a target detection special layer ROI Pooling; the technical scheme provided by the invention can effectively overcome the defect that the dynamic model cannot be accurately and effectively identified in the prior art.
Description
Technical Field
The invention relates to a dynamic model detection method, in particular to a dynamic model detection method based on an AI algorithm.
Background
The bright kitchen range enables customers of catering enterprises to visually see whether the operation of kitchen staff is standard, whether kitchen sanitation is qualified or not and whether some things affecting the kitchen environmental sanitation appear or not. Although many restaurants promote bright kitchen ranges, some catering enterprises have own small abacus and the implementation initiative is not enough. Many catering enterprises have simple and crude kitchen, the sanitation does not reach the standard, and the kitchen dares not to be disclosed, so the enthusiasm is not high, and the kitchen sanitation can not be effectively guaranteed.
Aiming at some animals in a kitchen, such as mice, a better observation means does not exist at present, the artificial observation cost is high, the animals cannot be identified in time, the existing camera is low in identification precision and feedback, and the condition of false alarm and missed alarm is caused.
Disclosure of Invention
Technical problem to be solved
Aiming at the defects in the prior art, the invention provides a dynamic model detection method based on an AI algorithm, which can effectively overcome the defect that the dynamic model cannot be accurately and effectively identified in the prior art.
(II) technical scheme
In order to achieve the purpose, the invention is realized by the following technical scheme:
a dynamic model detection method based on AI algorithm comprises the following steps:
s1, a system receives a picture evaluation request, acquires an input image from video access equipment, and carries out sharpening processing on the input image;
s2, loading a training model and a data set sample label, acquiring tensor in the training model, and converting an input image into a numpy array format;
s3, performing feature extraction on the input image by using the CNN convolutional layer;
s4, generating a suggestion window by using a Regional Proposal Network (RPN);
s5, mapping the suggestion window to the last convolution feature map of the CNN convolution layer, and enabling each ROI to generate a feature map with a fixed size through a target detection special layer ROI Pooling;
s6, carrying out classification regression and position regression;
and S7, carrying out position marking and type prediction on the dynamic model in the original input image.
Preferably, the sharpening processing on the input image in S1 includes: acquiring YUV data of an input image pixel, and normalizing the YUV data; performing neighborhood blurring on the Y component in the YUV data after normalization processing, calculating a Y component difference value before and after blurring, and calculating a Y component gain coefficient according to the Y component difference value; and performing clear enhancement on the input image by combining the Y component, the Y component difference and the Y component gain coefficient.
Preferably, the performing neighborhood blurring on the Y component in the normalized YUV data includes: and selecting adjacent NxN pixels by taking the current pixel as a central point, constructing a fuzzy matrix and a Y component matrix of the NxN pixels, wherein N is an odd number greater than 1, and performing neighborhood fuzzy according to data obtained by the fuzzy matrix and the Y component matrix.
Preferably, said calculating a Y component gain factor from the Y component difference comprises: and calculating a gain angle according to the Y component difference value, and obtaining a Y component gain coefficient according to the gain angle.
Preferably, the regional proposal network RPN generates 300 recommendation windows for each input image.
Preferably, said performing classification regression and location regression comprises: and performing combined training on the classification probability and the bounding box regression by using the detection classification probability Softmax Loss and the detection bounding box regression SmoothL1 Loss.
(III) advantageous effects
Compared with the prior art, the dynamic model detection method based on the AI algorithm, provided by the invention, firstly carries out sharpening processing on an input image to ensure the identification precision of a later dynamic model, then converts the input image into a numpy array format, carries out feature extraction on the input image by utilizing a CNN convolutional layer, generates an advice window by utilizing a Region Proposal Network (RPN), maps the advice window to the last layer of convolutional feature map of the CNN convolutional layer, enables each ROI to generate a feature map with a fixed size through the target detection special layer ROI Pooling, and carries out classification regression and position regression, thereby accurately and effectively marking the position of the dynamic model on the input image and predicting the type of the dynamic model.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. It is obvious that the drawings in the following description are only some embodiments of the invention, and that for a person skilled in the art, other drawings can be derived from them without inventive effort.
FIG. 1 is a schematic flow chart of the method of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
A dynamic model detection method based on AI algorithm, as shown in fig. 1, includes the following steps:
s1, a system receives a picture evaluation request, acquires an input image from video access equipment, and carries out sharpening processing on the input image;
s2, loading a training model and a data set sample label, acquiring tensor in the training model, and converting an input image into a numpy array format;
s3, performing feature extraction on the input image by using the CNN convolutional layer;
s4, generating a suggestion window by using a Regional Proposal Network (RPN);
s5, mapping the suggestion window to the last layer of convolution feature map of the CNN convolution layer, and enabling each ROI to generate a feature map with a fixed size through the target detection special layer ROI Pooling;
s6, carrying out classification regression and position regression;
and S7, carrying out position marking and type prediction on the dynamic model in the original input image.
The step of performing sharpening processing on the input image in the step S1 comprises the following steps: acquiring YUV data of an input image pixel, and normalizing the YUV data; performing neighborhood blurring on the Y component in the YUV data after normalization processing, calculating a Y component difference value before and after blurring, and calculating a Y component gain coefficient according to the Y component difference value; and performing clear enhancement on the input image by combining the Y component, the Y component difference and the Y component gain coefficient.
The neighborhood blurring of the Y component in the normalized YUV data comprises the following steps: selecting adjacent NxN pixels by taking the current pixel as a central point, constructing a fuzzy matrix and a Y component matrix of the NxN pixels, wherein N is an odd number larger than 1, and performing neighborhood fuzzy according to data obtained by fuzzy matrix and Y component matrix operation.
Calculating the Y component gain factor from the Y component difference comprises: and calculating a gain angle according to the Y component difference value, and obtaining a Y component gain coefficient according to the gain angle.
The regional proposal network RPN generates 300 recommendation windows for each input image.
Performing classification regression and location regression includes: and performing combined training on the classification probability and the border regression by using the detection classification probability Softmax Loss and the detection border regression SmoothL1 Loss.
The method comprises the steps of firstly carrying out sharpening processing on an input image to ensure the identification precision of a later dynamic model, then converting the input image into a numpy array format, utilizing a CNN convolutional layer to carry out feature extraction on the input image, utilizing a Region Proposal Network (RPN) to generate a suggestion window, mapping the suggestion window onto the last layer of convolutional feature map of the CNN convolutional layer, enabling each ROI of a region of interest to generate a feature map with a fixed size through a target detection special layer ROI Powing, and carrying out classification regression and position regression, thereby accurately and effectively marking the position of the dynamic model on the input image and predicting the type of the dynamic model.
The sharpening processing of the input image includes: acquiring YUV data of an input image pixel, and normalizing the YUV data; performing neighborhood blurring on the Y component in the YUV data after normalization processing, calculating a Y component difference value before and after blurring, and calculating a Y component gain coefficient according to the Y component difference value; and performing clear enhancement on the input image by combining the Y component, the Y component difference and the Y component gain coefficient.
The neighborhood blurring of the Y component in the normalized YUV data comprises the following steps: and selecting adjacent NxN pixels by taking the current pixel as a central point, constructing a fuzzy matrix and a Y component matrix of the NxN pixels, wherein N is an odd number greater than 1, and performing neighborhood fuzzy according to data obtained by the fuzzy matrix and the Y component matrix.
Calculating the Y component gain factor from the Y component difference comprises: and calculating a gain angle according to the Y component difference value, and obtaining a Y component gain coefficient according to the gain angle.
Performing classification regression and location regression includes: and performing combined training on the classification probability and the bounding box regression by using the detection classification probability Softmax Loss and the detection bounding box regression Smooth L1 Loss.
It is worth noting that the technical scheme of the application is not limited to be applied to detecting rats in a kitchen, and the technical scheme of the application can be applied to all environments for dynamic model detection.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit and scope of the corresponding technical solutions.
Claims (6)
1. A dynamic model detection method based on AI algorithm is characterized in that: the method comprises the following steps:
s1, a system receives a picture evaluation request, acquires an input image from video access equipment, and carries out sharpening processing on the input image;
s2, loading a training model and a data set sample label, acquiring tensor in the training model, and converting an input image into a numpy array format;
s3, performing feature extraction on the input image by using the CNN convolutional layer;
s4, generating a suggestion window by using a Regional Proposal Network (RPN);
s5, mapping the suggestion window to the last layer of convolution feature map of the CNN convolution layer, and enabling each ROI to generate a feature map with a fixed size through a target detection special layer ROIPooling;
s6, carrying out classification regression and position regression;
and S7, carrying out position marking and type prediction on the dynamic model in the original input image.
2. The AI-algorithm-based dynamic model detection method of claim 1, characterized in that: the sharpening processing of the input image in the S1 comprises the following steps: acquiring YUV data of an input image pixel, and normalizing the YUV data; performing neighborhood blurring on the Y component in the YUV data after normalization processing, calculating a Y component difference value before and after blurring, and calculating a Y component gain coefficient according to the Y component difference value; and performing clear enhancement on the input image by combining the Y component, the Y component difference and the Y component gain coefficient.
3. The AI algorithm-based dynamic model detection method of claim 2, characterized in that: the neighborhood blurring of the Y component in the normalized YUV data comprises: selecting adjacent NxN pixels by taking the current pixel as a central point, constructing a fuzzy matrix and a Y component matrix of the NxN pixels, wherein N is an odd number larger than 1, and performing neighborhood fuzzy according to data obtained by fuzzy matrix and Y component matrix operation.
4. The AI-algorithm-based dynamic model detection method of claim 2, characterized in that: the calculating the Y component gain factor according to the Y component difference value includes: and calculating a gain angle according to the Y component difference value, and obtaining a Y component gain coefficient according to the gain angle.
5. The AI algorithm-based dynamic model detection method of claim 1, wherein: the regional proposal network RPN generates 300 recommendation windows for each input image.
6. The AI algorithm-based dynamic model detection method of claim 1, wherein: performing classification regression and location regression includes: and performing combined training on the classification probability and the border regression by using the detection classification probability Softmax Loss and the detection border regression Smooth L1 Loss.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910860225.7A CN110705604B (en) | 2019-09-11 | 2019-09-11 | AI algorithm-based dynamic model detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910860225.7A CN110705604B (en) | 2019-09-11 | 2019-09-11 | AI algorithm-based dynamic model detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110705604A CN110705604A (en) | 2020-01-17 |
CN110705604B true CN110705604B (en) | 2022-11-29 |
Family
ID=69195100
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910860225.7A Active CN110705604B (en) | 2019-09-11 | 2019-09-11 | AI algorithm-based dynamic model detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110705604B (en) |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10289938B1 (en) * | 2017-05-16 | 2019-05-14 | State Farm Mutual Automobile Insurance Company | Systems and methods regarding image distification and prediction models |
CN110210463A (en) * | 2019-07-03 | 2019-09-06 | 中国人民解放军海军航空大学 | Radar target image detecting method based on Precise ROI-Faster R-CNN |
-
2019
- 2019-09-11 CN CN201910860225.7A patent/CN110705604B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10289938B1 (en) * | 2017-05-16 | 2019-05-14 | State Farm Mutual Automobile Insurance Company | Systems and methods regarding image distification and prediction models |
CN110210463A (en) * | 2019-07-03 | 2019-09-06 | 中国人民解放军海军航空大学 | Radar target image detecting method based on Precise ROI-Faster R-CNN |
Non-Patent Citations (1)
Title |
---|
基于深度学习的交通标志检测方法研究;武林秀等;《大连民族大学学报》;20180915(第05期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN110705604A (en) | 2020-01-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11392792B2 (en) | Method and apparatus for generating vehicle damage information | |
US11915447B2 (en) | Audio acquisition device positioning method and apparatus, and speaker recognition method and system | |
TWI716012B (en) | Sample labeling method, device, storage medium and computing equipment, damage category identification method and device | |
CN112613569B (en) | Image recognition method, training method and device for image classification model | |
CN112989995B (en) | Text detection method and device and electronic equipment | |
CN113436100A (en) | Method, apparatus, device, medium and product for repairing video | |
WO2020259416A1 (en) | Image collection control method and apparatus, electronic device, and storage medium | |
CN113222921A (en) | Image processing method and system | |
CN114359932B (en) | Text detection method, text recognition method and device | |
CN113988222A (en) | Forest fire detection and identification method based on fast-RCNN | |
CN110705604B (en) | AI algorithm-based dynamic model detection method | |
CN112113638B (en) | Water meter function self-checking device and method | |
CN113888438A (en) | Image processing method, device and storage medium | |
Salih et al. | Adaptive local exposure based region determination for non-uniform illumination and low contrast images | |
CN116071651B (en) | Voltage equalizing field identification method and device, storage medium and terminal | |
Pratomo et al. | Parking detection system using background subtraction and HSV color segmentation | |
CN116052097A (en) | Map element detection method and device, electronic equipment and storage medium | |
CN114419070A (en) | Image scene segmentation method, device, equipment and storage medium | |
CN114757941A (en) | Transformer substation equipment defect identification method and device, electronic equipment and storage medium | |
CN112348823A (en) | Object-oriented high-resolution remote sensing image segmentation algorithm | |
WO2021189460A1 (en) | Image processing method and apparatus, and movable platform | |
CN113449617A (en) | Track safety detection method, system, device and storage medium | |
KR20230043419A (en) | Ai-based building defect inspection system | |
CN113505860A (en) | Screening method and device for blind area detection training set, server and storage medium | |
CN112153320A (en) | Method and device for measuring size of article, electronic equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |