CN-121686252-B - Intelligent inspection method and system based on AI image recognition

CN121686252BCN 121686252 BCN121686252 BCN 121686252BCN-121686252-B

Abstract

The application provides an intelligent inspection method and system based on AI image recognition, and relates to the technical field of image recognition analysis. The method comprises the steps of firstly obtaining visible light images and thermal imaging images of a patrol area and quality evaluation data, secondly dynamically selecting and combining processing paths from a plurality of preset processing paths, simultaneously carrying out feature improvement on the visible light images to generate a standardized feature map, then combining the standardized feature map with a temperature feature map to generate a combined feature map, finally inputting the combined feature map into a pre-trained image classification model, and outputting a judgment result of food manufacturing process compliance. The technical scheme provided by the application improves the judgment performance of intelligent inspection on the process compliance in a complex real environment through a technical chain with the advantages of perceived quality, optimized characteristics, accurate fusion and stable decision loop-to-loop.

Inventors

LEI YUYUN
XUE JUNHAO
LI XIANG

Assignees

中易云(北京)物联网科技集团有限公司

Dates

Publication Date: 20260512
Application Date: 20260211

Claims (8)

1. An intelligent inspection method based on AI image recognition is characterized by comprising the following steps: Obtaining visible light images and thermal imaging images of the inspection area and quality evaluation data; Dynamically selecting and combining processing paths from a plurality of preset processing paths based on the quality evaluation data, and performing feature improvement on the visible light image according to the combined processing paths to generate a standardized feature map; Combining the standardized feature map with a temperature feature map in the thermal imaging image at the same time to generate a combined feature map; Inputting the combined feature map into a pre-trained image classification model, and outputting a judging result of food manufacturing process compliance based on the combined feature map; The method for acquiring the visible light image and the thermal imaging image of the inspection area and the quality evaluation data comprises the following steps: synchronously acquiring a visible light image containing scene details reflecting the inspection area and a thermal imaging image reflecting temperature distribution of the inspection area through an imaging device which is arranged in the inspection area in advance; the visible light image is analyzed to obtain global illumination parameters and regional definition parameters describing the visible light image; Extracting temperature fluctuation characteristics representing the stability of temperature reading in the thermal imaging image by analyzing the thermal imaging image; integrating the global illumination parameter, the regional definition parameter and the temperature fluctuation characteristic to generate quality evaluation data; the step of dynamically selecting and combining processing paths from a plurality of preset processing paths based on the quality evaluation data, and performing feature improvement on the visible light image according to the combined processing paths to generate a standardized feature map, includes: Comparing the global illumination parameter in the quality evaluation data with a plurality of preset illumination intensity threshold intervals, and selecting a target image brightness adjustment path from a preset image brightness adjustment path sequence according to the illumination intensity threshold interval where the global illumination parameter is located; Comparing the regional definition parameters in the quality evaluation data with a preset definition judgment threshold, and if the regional definition parameters are lower than the definition judgment threshold, starting a preset image sharpening enhancement path; Comparing the temperature fluctuation characteristic in the quality evaluation data with a preset temperature stability threshold value, and selecting a target noise suppression path from a preset noise suppression path sequence according to the change trend of the temperature fluctuation characteristic relative to the temperature stability threshold value; combining the target image brightness adjustment path, the preset image sharpening enhancement path and the target noise suppression path into an image processing path according to a preset execution sequence, wherein whether the preset image sharpening enhancement path is started or not is determined according to a judging result; according to the processing mode defined by each path in the image processing paths, image brightness adjustment, image sharpening enhancement and noise suppression operations are sequentially carried out on the visible light image, so that a processed visible light image is obtained; And executing feature extraction operation on the processed visible light image to generate a standardized feature map, wherein the standardized feature map is used for describing the scene details of the inspection area.
2. The method of claim 1, wherein combining the normalized signature with a temperature signature from the thermographic image at the same time to generate a combined signature comprises: dividing the thermal imaging image into a plurality of temperature-related areas according to the continuity and the difference of temperature distribution in the thermal imaging image, wherein each temperature-related area comprises a pixel set with similar temperature characteristics, wherein the similar temperature characteristics refer to the property that the temperature values of a plurality of pixel points in the thermal imaging image are close in value or smaller in change amplitude, the property is used for defining a characteristic that the temperature distribution in the spatially continuous area has higher consistency, and clustering and boundary division are carried out according to the temperature difference between the pixels through an image dividing algorithm; Calculating a characteristic temperature value representing a temperature level of the temperature-related region; Establishing a mapping relation between the position of each temperature association region in the thermal imaging image and a corresponding position region in the standardized feature map; according to the mapping relation, assigning the characteristic temperature value corresponding to each temperature association region to each characteristic point in the corresponding position region in the standardized characteristic map; Taking each feature point in the standardized feature map as a reference, correlating the detail feature information contained in the feature points with the given feature temperature value to form target feature points carrying detail features and temperature features; All target feature points are aggregated to form a combined feature map.
3. The method of claim 1, wherein inputting the combined feature map into a pre-trained image classification model, outputting a determination of food preparation process compliance based on the combined feature map, comprises: inputting the combined feature map into a pre-trained image classification model; Defining a plurality of process semantic levels in the image classification model according to different process stages associated with a food production process; Dividing the target feature points into a corresponding process semantic category in each process semantic hierarchy based on the detail feature information and the feature temperature value carried by each target feature point in the combined feature map, and determining a first confidence coefficient of the target feature points belonging to the process semantic category; Calculating the second confidence coefficient of each process semantic category in the process semantic level according to the first confidence coefficient of all target feature points of different process semantic categories in the process semantic level; determining a process semantic category with the highest second confidence in each process semantic level as a first judging result of the process semantic level; Synthesizing the first judging results of all the process semantic levels to generate a process compliance judging vector; based on the process compliance determination vector, a final determination result for food production process compliance is output.
4. The method of claim 2, wherein calculating a characteristic temperature value representative of a temperature level of the temperature-dependent region comprises: acquiring a pixel temperature value set of all pixel points in the temperature-related area; In the pixel temperature value set, all pixel temperature values are arranged according to the sequence from the highest temperature to the lowest temperature, so as to obtain a pixel temperature value sequence; Respectively removing pixel temperature values of a first preset proportion from the head end and the tail end of the pixel temperature value sequence to obtain a pixel temperature value subset of the middle region; Calculating the arithmetic average value of all pixel temperature values in the pixel temperature value subset of the middle area to obtain an initial characteristic temperature value; determining all pixel temperature values, of which the difference value from the initial characteristic temperature value is within a second preset range, in the pixel temperature value subset of the middle area to form a core temperature subset; and calculating a characteristic temperature value based on the distribution mode of all pixel temperature values in the core temperature subset.
5. A method according to claim 3, wherein calculating a second confidence level for each process semantic category at the process semantic level based on the first confidence levels of all target feature points partitioned into different process semantic categories within the process semantic level comprises: counting the number of all target feature points divided into the same process semantic category in the process semantic hierarchy; accumulating the first confidence coefficient of all the target feature points belonging to the same process semantic category to obtain the accumulated confidence coefficient sum of the process semantic category; calculating the ratio of the accumulated confidence coefficient sum to the number to obtain the initial level confidence coefficient of the process semantic category; determining a variation range of a first confidence coefficient of all target feature points contained in the process semantic category; selecting a target confidence coefficient correction strategy from a preset confidence coefficient correction strategy set according to the change range; And carrying out numerical adjustment on the initial level confidence coefficient by using a target confidence coefficient correction strategy so as to obtain a second confidence coefficient of the process semantic category at the process semantic level.
6. An intelligent inspection system based on AI image recognition, which is applied to the intelligent inspection method based on AI image recognition of any one of claims 1-5, and is characterized by comprising: The acquisition module is used for acquiring visible light images and thermal imaging images of the inspection area and quality evaluation data; The improvement module is used for dynamically selecting and combining processing paths from a plurality of preset processing paths based on the quality evaluation data, and carrying out feature improvement on the visible light image according to the combined processing paths so as to generate a standardized feature map; the combination module is used for combining the standardized feature map with the temperature feature map in the thermal imaging image at the same moment to generate a combined feature map; And the output module is used for inputting the combined feature map into a pre-trained image classification model and outputting a judging result of the food manufacturing process compliance based on the combined feature map.
7. The computing device is characterized by comprising a processing component and a storage component, wherein the storage component stores one or more computer instructions, and the one or more computer instructions are used for being invoked and executed by the processing component to realize the intelligent inspection method based on AI image recognition according to any one of claims 1-5.
8. A computer storage medium, wherein a computer program is stored, and when the computer program is executed by a computer, the intelligent inspection method based on AI image recognition is realized according to any one of claims 1-5.

Description

Intelligent inspection method and system based on AI image recognition Technical Field The application relates to the technical field of image recognition analysis, in particular to an intelligent inspection method and system based on AI image recognition. Background Along with the continuous improvement of the requirements of the food and beverage and food processing industry on the standardization of production flow and the management and control of food safety, an intelligent inspection technology capable of automatically and precisely monitoring key parameters of a food manufacturing process is urgently needed to replace the traditional manual inspection which is strong in subjectivity, low in efficiency and difficult to continuously. At present, the prior art adopts an intelligent monitoring method based on multi-mode image fusion, the method registers and fuses a visible light image and a thermal imaging image by synchronously acquiring the two images to form comprehensive data containing appearance textures and temperature information, and the comprehensive data is directly input into a pre-trained image classification model to judge whether the current process state meets a preset standard or not. However, the image processing flow and the feature fusion mode of the method are usually fixed, the method is difficult to adapt to complex and changeable real industrial field environments, and when field illumination is changed severely, a camera is slightly out of focus to cause image blurring, or a thermal imaging sensor is subjected to short-term interference, a fixed image preprocessing and feature extraction strategy is difficult to generate stable and reliable feature expression to cause the reduction of the data quality of an input classification model, and the direct influence of the image quality on the fusion effect and the differential influence of different quality dimensions on final decisions are often ignored during fusion, so that the judgment accuracy and the robustness are insufficient under non-ideal conditions, and false alarm or missing report easily occurs. Disclosure of Invention The application provides an intelligent inspection method and system based on AI image recognition, which are used for solving the problem that the reliability of process consistency judgment is not high in a complex environment because a process cannot be dynamically optimized according to real-time image quality and a fusion strategy in the prior art. In a first aspect, the present application provides an intelligent inspection method based on AI image recognition, including: Obtaining visible light images and thermal imaging images of the inspection area and quality evaluation data; Dynamically selecting and combining processing paths from a plurality of preset processing paths based on the quality evaluation data, and performing feature improvement on the visible light image according to the combined processing paths to generate a standardized feature map; Combining the standardized feature map with a temperature feature map in the thermal imaging image at the same time to generate a combined feature map; inputting the combined feature map into a pre-trained image classification model, and outputting a judging result of the food manufacturing process compliance based on the combined feature map. Optionally, acquiring the visible light image and the thermal imaging image of the inspection area and the quality evaluation data includes: synchronously acquiring a visible light image containing scene details reflecting the inspection area and a thermal imaging image reflecting temperature distribution of the inspection area through an imaging device which is arranged in the inspection area in advance; the visible light image is analyzed to obtain global illumination parameters and regional definition parameters describing the visible light image; Extracting temperature fluctuation characteristics representing the stability of temperature reading in the thermal imaging image by analyzing the thermal imaging image; and integrating the global illumination parameter, the regional definition parameter and the temperature fluctuation characteristic to generate quality evaluation data. Optionally, based on the quality evaluation data, dynamically selecting and combining processing paths from a plurality of preset processing paths, and performing feature improvement on the visible light image according to the combined processing paths to generate a standardized feature map, including: Comparing the global illumination parameter in the quality evaluation data with a plurality of preset illumination intensity threshold intervals, and selecting a target image brightness adjustment path from a preset image brightness adjustment path sequence according to the illumination intensity threshold interval where the global illumination parameter is located; Comparing the regional definition parameters in the quality evaluatio