CN-122023891-A - Visual identification-based method and system for detecting appearance defects of auxiliary materials of cigarettes

CN122023891ACN 122023891 ACN122023891 ACN 122023891ACN-122023891-A

Abstract

The invention relates to the technical field of image appearance defect detection, in particular to a method and a system for detecting the appearance defect of a cigarette auxiliary material based on visual identification, which are characterized in that a high-precision image acquisition system is built to acquire a cigarette auxiliary material image and working parameters of a light source in real time, the cigarette auxiliary material image is uniformly divided into a plurality of auxiliary material area blocks, the area characteristics of the auxiliary material area blocks are identified to distinguish suspected defect area blocks, the area stability is analyzed based on the suspected defect area blocks, the method is used for screening the suspected region blocks and the defect region blocks, regulating and controlling the first light source working parameters as the second light source working parameters within the preset working parameter range, acquiring corresponding first cigarette auxiliary material images according to the second light source working parameters, carrying out double defect analysis on the suspected region blocks to determine the defect region blocks, extracting appearance defect characteristics based on the defect region blocks, constructing a defect detection model based on deep learning, reducing the probability of missing detection and false detection, and realizing intelligent classification and identification of the cigarette auxiliary material image defects.

Inventors

YAN YONG
LIU JIANGHUA
FANG LI
YU QI
WANG YIBIN
LIU CHENGJUN
YANG FEIYING
XU WENWU
LI BIN
ZHAO GUANGWEN
LI JINGZU
FAN SHENGXING
ZHONG YU
ZHANG FU
ZHAO ZIQI
LIANG JIAJI
LI SHUAI

Assignees

南京焦耳科技有限责任公司

Dates

Publication Date: 20260512
Application Date: 20260122

Claims (10)

1. The visual identification-based method for detecting the appearance defects of the auxiliary materials of the cigarettes is characterized by comprising the following steps of: Acquiring a cigarette auxiliary material image in real time as a first cigarette auxiliary material image by building a high-precision image acquisition system, acquiring a light source working parameter as a first light source working parameter during image acquisition in real time, acquiring a preset standard cigarette auxiliary material image as a second cigarette auxiliary material image, and respectively preprocessing the cigarette auxiliary material image and the first light source working parameter; The method comprises the steps of uniformly dividing a first cigarette auxiliary material image into a plurality of auxiliary material area blocks, identifying the area characteristics of the auxiliary material area blocks to distinguish suspected defect area blocks, analyzing the area stability degree based on the suspected defect area blocks, and screening suspected area blocks and defect area blocks; Acquiring a first light source working parameter of an suspicious region block, regulating and controlling the first light source working parameter as a second light source working parameter within a preset working parameter range, acquiring a corresponding first cigarette auxiliary material image according to the second light source working parameter, and carrying out double defect analysis on the suspicious region block to determine a defect region block; And extracting appearance defect characteristics based on the defect area blocks, constructing a defect detection model based on deep learning, and realizing intelligent classification and identification of the image defects of the auxiliary materials of the cigarettes.
2. The visual identification-based method for detecting the appearance defects of the auxiliary materials of the cigarettes according to claim 1, wherein the method is characterized in that the first cigarette auxiliary material image is uniformly divided into a plurality of auxiliary material area blocks, the area characteristics of the auxiliary material area blocks are identified to distinguish suspected defect area blocks, the area stability is analyzed based on the suspected defect area blocks, and the method is used for screening suspected area blocks and defect area blocks and comprises the following specific steps: uniformly dividing a first cigarette auxiliary material image in a row direction and a column direction according to preset block size parameters to form a plurality of auxiliary material area blocks, wherein each auxiliary material area block corresponds to a local area of the surface of a cigarette auxiliary material; Extracting region characteristics including region gray value characteristics and region texture characteristics for each auxiliary material region block respectively; Extracting regional gray value characteristics of the regional blocks, calculating regional contrast coefficients, and identifying suspected defective regional blocks based on the regional contrast coefficients; And extracting region texture characteristics from the suspected defective region blocks, identifying whether the suspected defective region blocks are in a high fluctuation state or a low fluctuation state, constructing a region stability evaluation model for the suspected defective region blocks in the high fluctuation state, calculating the region stability degree, and judging the first heavy defect based on the region stability degree.
3. The visual identification-based method for detecting the appearance defects of the auxiliary materials of the cigarettes according to claim 2, wherein the method is characterized in that the regional contrast coefficient is calculated by extracting regional gray value characteristics of regional blocks, and the suspected defective regional blocks are identified based on the regional contrast coefficient, and specifically comprises the following steps: comparing the regional contrast coefficient with a preset contrast threshold, if the regional contrast coefficient is larger than or equal to the preset contrast threshold, judging the regional block corresponding to the regional contrast coefficient as a suspected defective regional block, and if the regional contrast coefficient is smaller than the preset contrast threshold, judging the regional block corresponding to the regional contrast coefficient as a non-defective regional block.
4. The visual recognition-based method for detecting an appearance defect of a cigarette accessory of claim 2, wherein the regional texture features include texture direction distribution features and texture consistency features.
5. The visual identification-based method for detecting the appearance defects of the cigarette auxiliary materials according to claim 4, wherein the texture direction distribution characteristics are specifically that gradient calculation is carried out on gray images corresponding to suspected defective area blocks, horizontal direction gradients and vertical direction gradients are obtained, and texture direction angles of pixels in the area blocks are calculated based on the horizontal direction gradients and the vertical direction gradient components and serve as the texture direction distribution characteristics.
6. The visual identification-based cigarette accessory appearance defect detection method of claim 4, wherein the texture consistency characteristics are specifically that local texture descriptors are calculated in a suspected defect area block and used for describing texture modes of pixel neighborhoods, statistical analysis is carried out on the texture descriptors to obtain texture statistical characteristics in the area, texture statistical characteristic difference values between adjacent pixels are calculated according to the spatial fluctuation condition of the texture statistical characteristics in the suspected defect area block, and the difference value of the maximum value and the minimum value in the texture statistical characteristic difference values is extracted to be used as the texture consistency characteristics of corresponding pixels.
7. The visual identification-based method for detecting the appearance defects of the auxiliary materials of the cigarettes according to claim 3, wherein the method for detecting the appearance defects of the auxiliary materials of the cigarettes is characterized by extracting regional texture features from the suspected defective regional blocks and identifying whether the suspected defective regional blocks are in a high fluctuation state or a low fluctuation state, and comprises the following specific steps: Extracting texture consistency characteristics in texture characteristics of the region, calculating the average value of the texture consistency characteristics corresponding to each pixel in the suspected defect region block as a texture fluctuation identification index, comparing the texture fluctuation identification index with a preset texture fluctuation identification threshold, if the texture fluctuation identification index is greater than or equal to the preset texture fluctuation identification threshold, enabling the suspected defect region block to be in a high fluctuation state, and if the texture fluctuation identification index is smaller than the preset texture fluctuation identification threshold, enabling the suspected defect region block to be in a low fluctuation state.
8. The visual identification-based method for detecting the appearance defects of the auxiliary materials of the cigarettes according to claim 2, wherein for the suspected defective area blocks in a high fluctuation state, an area stability evaluation model is constructed to calculate the area stability degree, and the first heavy defect judgment is performed based on the area stability degree, and the specific steps are as follows: For a suspected defect area block in a high fluctuation state, respectively extracting texture direction distribution characteristics and texture consistency characteristics corresponding to each pixel in the suspected defect area block, taking the average value of the texture direction distribution characteristics corresponding to each pixel in the suspected defect area block as a first stability factor, taking the average value of the texture consistency characteristics as a second stability factor, and carrying out dynamic weighted combination based on the first stability factor and the second stability factor to construct an area stability evaluation model to calculate the area stability degree; And judging the first heavy defect based on the area stability, comparing the area stability with a preset stability threshold, marking the corresponding suspected defective area block as a suspected area block if the area stability is greater than or equal to the preset stability threshold, and marking the corresponding suspected defective area block as a defective area block if the area stability is less than the preset stability threshold.
9. The visual identification-based method for detecting the appearance defects of the auxiliary materials of the cigarettes, which is disclosed in claim 1, is characterized by comprising the following specific steps of: acquiring a first light source working parameter of the suspicious region block, and regulating and controlling the first light source working parameter as a second light source working parameter within a preset working parameter range; Analyzing and calculating the area stability degree based on the suspicious area block under the working parameters of the second light source to serve as a second area stability degree, and screening and determining the non-defective area block and the defective area block according to the second area stability degree of the suspicious area block; Comparing the second area stability of the suspected area block with a preset stability threshold, calculating the difference value between the second area stability and the area stability if the second area stability is greater than or equal to the preset stability threshold, marking the corresponding suspected defective area block as a non-defective area block if the difference value is less than the preset difference threshold, and marking the corresponding suspected defective area block as a defective area block if the difference value is less than the preset difference threshold.
10. The visual identification-based cigarette accessory appearance defect detection system is applied to the visual identification-based cigarette accessory appearance defect detection method according to any one of claims 1-9, and is characterized by comprising an image data acquisition processing module, a region block division and suspected defect screening module, a region stability analysis and type judgment module and an intelligent defect feature extraction and identification module; The image data acquisition processing module is used for acquiring a cigarette auxiliary material image in real time to serve as a first cigarette auxiliary material image by building a high-precision image acquisition system, acquiring a light source working parameter as a first light source working parameter during image acquisition in real time, acquiring a preset standard cigarette auxiliary material image as a second cigarette auxiliary material image, and preprocessing the cigarette auxiliary material image and the first light source working parameter respectively; The regional block division and suspected defect screening module is used for identifying regional features of the auxiliary material regional blocks to distinguish suspected defect regional blocks by uniformly dividing the first cigarette auxiliary material image into a plurality of auxiliary material regional blocks, analyzing regional stability based on the suspected defect regional blocks and screening suspected storage regional blocks and defect regional blocks; The area stability analysis and type judgment module is used for acquiring a first light source working parameter of the suspicious area block, regulating and controlling the first light source working parameter as a second light source working parameter within a preset working parameter range, acquiring a corresponding first cigarette auxiliary material image according to the second light source working parameter, and carrying out double defect analysis on the suspicious area block to determine a defect area block; the intelligent defect feature extraction and identification module is used for extracting appearance defect features based on the defect area blocks, constructing a defect detection model based on deep learning, and realizing intelligent classification and identification of the image defects of the auxiliary materials of the cigarettes.

Description

Visual identification-based method and system for detecting appearance defects of auxiliary materials of cigarettes Technical Field The invention relates to the technical field of image appearance defect detection, in particular to a method and a system for detecting appearance defects of auxiliary materials of cigarettes based on visual identification. Background In the production process of a cigarette factory, the cigarette auxiliary materials are usually purchased by an external provider, the appearance quality of the cigarette auxiliary materials directly influences the compliance, brand image and anti-counterfeiting function of the cigarette products, so that strict detection needs to be carried out before the auxiliary materials enter the factory or come on line, and complex texture auxiliary materials such as gold stamping tipping paper, holographic anti-counterfeiting trademark paper and high-grade auxiliary materials printed by metal ink exist on the cigarette auxiliary materials, and the surface of the auxiliary materials generally has strong specular reflection characteristics. When the auxiliary material passes through the detection station on the high-speed production line, the illumination light source with fixed angle and fixed brightness easily forms strong reflection or overexposure bright spots in a local area, so that the appearance defect originally existing in the area is covered by the highlight reflection signal in the image. Such reflective regions often appear as gray saturation or texture detail loss states in imaging results, resulting in failure to extract effective features by subsequent defect recognition algorithms based on threshold values, texture features or deep learning models, thereby generating a miss detection phenomenon. Because the complex texture auxiliary material has dynamic texture characteristics changing along with the observation angle, the reflection intensity difference of different positions is obvious in the same auxiliary material image. In the prior art, uniform illumination parameters and exposure parameters are generally adopted for acquisition, imaging quality of a high-reflection area and an imaging quality of a low-reflection area are difficult to be simultaneously considered, when the system integrally reduces illumination or exposure for avoiding overexposure, defect contrast of a non-reflection area is reduced, so that low-significance appearance defects such as micro chromatic aberration, hairline scratches and the like are further submerged by background textures, and thus the problem of imaging unbalance of 'local overexposure and overall underexposure' is integrally formed, and enterprises are forced to repeatedly adjust angles, brightness or detection thresholds of light sources through manual experience, so that compromise effects are achieved among auxiliary materials of different batches and different specifications. However, the method is difficult to adapt to high-speed continuous production conditions, is highly sensitive to equipment states and ambient light changes, and has the defects of poor stability of detection results and obvious fluctuation of omission rate. Therefore, in the complex texture auxiliary material scene, the reflection interference becomes one of key problems for restricting the detection precision and the engineering usability of the existing appearance defect detection technology. Therefore, it is necessary to provide a method and a system for detecting appearance defects of auxiliary materials of cigarettes based on visual recognition to solve the above technical problems, and in order to solve the above problems, a technical scheme is provided. Disclosure of Invention In order to overcome the defects in the prior art, the invention provides a visual identification-based method and a visual identification-based system for detecting the appearance defects of a cigarette auxiliary material, which are used for solving the problems that under the condition of high-speed production, the cigarette auxiliary material with complex texture is subjected to local overexposure and texture information loss due to strong specular reflection, different reflection areas are difficult to consider when illumination parameters are fixed, so that the appearance defects are easy to be covered by reflection and missed detection is generated, and the stability and the accuracy of the existing detection method are restricted. In order to achieve the above purpose, the present invention provides the following technical solutions: The visual identification-based method for detecting the appearance defects of the auxiliary materials of the cigarettes comprises the following steps: Acquiring a cigarette auxiliary material image in real time as a first cigarette auxiliary material image by building a high-precision image acquisition system, acquiring a light source working parameter as a first light source working parameter during image a