Industrial detection system of SWIT6 millimeter lens based on neural network

Abstract

The invention relates to the technical field of optical lenses, in particular to an industrial detection system of a SWIT6 millimeter lens based on a neural network, which comprises a lens industrial detection center, a basic detection unit, a scanning analysis unit, a mechanical analysis unit, an optical analysis unit, an evaluation feedback unit and an early warning feedback unit, according to the method, analysis is performed preliminarily from the angle of the basic size of the target SWIT6 millimeter lens, on one hand, the detection precision is improved, on the other hand, sorting and management of the abnormal target SWIT6 millimeter lens are facilitated, detection is further performed from the two angles of mechanical performance and optical performance through an information feedback mode, and the detection accuracy is improved. According to the method and the device, the target SWIT6 millimeter lens is detected to judge whether the target SWIT6 millimeter lens is qualified or not, so that the target SWIT6 millimeter lens is further sorted and managed, and meanwhile, the problems of the target SWIT6 millimeter lens in the current batch are checked along with the summarized analysis of the detection output signals, so that the production qualification rate is increased.

Description

Technical Field

[0001] The invention relates to the technical field of optical lenses, and in particular to an industrial detection system for SWIR 6 mm lenses based on a neural network. Background Art

[0002] In today's era of booming science and technology, short-wave infrared (SWIR) imaging technology is gradually penetrating into various key fields, making the demand for high-performance SWIR lenses more urgent. In the industrial manufacturing inspection process, especially in high-precision industries such as semiconductors and precision optical component processing, the detection accuracy of internal defects in materials and tiny defects in components is extremely high. SWIR lenses with large target areas can cover a larger area of ​​workpieces in one imaging. Combined with the short-focus and wide-angle characteristics, they can perform all-round and blind-angle observation of parts with complex shapes. Among them, the SWIR6mm lens, with its unique optical characteristics and industrial adaptability, shows the advantages of high precision and high efficiency in the detection scene. However, in the existing technology, it is impossible to perform high-precision detection on the size, mechanical and optical performance of the SWIR6mm lens. On the one hand, it is not conducive to the sorting and management of the SWIR6mm lens, reducing the management efficiency of the SWIR6mm lens. On the other hand, it is not conducive to the sorting and management of production defects, reducing the production efficiency of the SWIR6mm lens. In view of the above technical defects, a solution is now proposed. Summary of the invention

[0003] The purpose of the present invention is to provide an industrial detection system for SWIR6mm lenses based on neural networks to solve the technical defects mentioned above. The present invention initially analyzes from the perspective of the basic size of the target SWIR6mm lens, and detects the basic size of the target SWIR6mm lens through two points of neural network and laser scanning technology. On the one hand, it helps to improve the detection accuracy, and on the other hand, it helps to sort and manage abnormal target SWIR6mm lenses. Through information feedback, the target SWIR6mm lens is further detected from the two perspectives of mechanical properties and optical properties to determine whether the target SWIR6mm lens is qualified, so as to further sort and manage the target SWIR6mm lens. At the same time, the detection output signal is summarized and analyzed to check the problems existing in the current batch of target SWIR6mm lenses to improve the production qualification rate of the target SWIR6mm lenses.

[0004] The object of the present invention can be achieved by the following technical solutions: an industrial detection system for SWIR 6 mm lens based on a neural network, comprising a lens industrial detection center, a basic detection unit, a scanning analysis unit, a mechanical analysis unit, an optical analysis unit, an evaluation feedback unit and an early warning feedback unit; The lens industrial inspection center is used to collect the appearance feature image of the target SWIR6 mm lens, and send the appearance feature image to the basic inspection unit for basic size inspection and analysis to obtain a normal signal or an abnormal signal. The scanning analysis unit is used to compare and analyze the collected surface size information of the target SWIR6 mm lens one by one to obtain a consistent signal or a difference signal, and interactively output and analyze the consistent signal, the difference signal, the normal signal and the abnormal signal to obtain a standard signal or a defect signal. When the standard signal is generated, the mechanical analysis unit is used to perform mechanical quality detection and analysis on the collected mechanical detection data to obtain a mechanical qualified signal or a mechanical unqualified signal, and the optical analysis unit is used to perform optical performance detection and analysis on the collected optical detection data to obtain an optical qualified signal or an optical deviation signal; The evaluation feedback unit is used to retrieve the mechanical qualified signal, mechanical unqualified signal, optical qualified signal and optical deviation signal for dual performance detection and analysis to obtain a quality output signal or a defect output signal. At the same time, the batch quality deviation rate is obtained through dual performance detection and analysis, and the batch quality deviation rate is compared and analyzed to obtain a batch quality signal or a batch risk signal.

[0005] Preferably, the basic size detection and analysis process is as follows: the detection period of the target SWIR6 mm lens is collected, and the detection period of the target SWIR6 mm lens is set as a time threshold, multiple groups of appearance feature images of the target SWIR6 mm lens corresponding to qualified or unqualified sizes are obtained, an appearance feature image set is constructed based on the multiple groups of appearance feature images of the target SWIR6 mm lens, and the appearance feature image set is preprocessed, and the preprocessed appearance feature image set is trained by AI deep learning and machine vision technology, and finally a size detection model is obtained; The surface size image of the target SWIR6mm lens is obtained, and the surface size image includes the lens flange surface feature image and the threaded interface feature image. The surface size image is input into the size detection model, and the output information of the size detection model is obtained. The output information includes qualified surface size and unqualified surface size. The output information corresponding to the surface size image input into the size detection model is that the number of unqualified surface sizes is set as the size measurement index, and the size measurement index is compared and analyzed to obtain a normal signal or an abnormal signal.

[0006] Preferably, the one-by-one comparison and analysis process is as follows: the surface dimension information of the target SWIR6mm lens within the time threshold is obtained through laser scanning technology, the surface dimension information includes the lens flange surface dimension and the threaded interface dimension, and the surface dimension information is compared and analyzed one by one with the standard surface dimension information on the design drawing of the target SWIR6mm lens, and the output result of the comparison and analysis is obtained, the output result includes a consistent result and a different result, if the output result of the comparison and analysis is a consistent result, a consistent signal is generated, if the output result of the comparison and analysis is a different result, a different signal is generated.

[0007] Preferably, the mechanical quality inspection and analysis process is as follows: mechanical inspection data of the target SWIR6mm lens within a time threshold is obtained, the mechanical inspection data including autofocus performance values ​​and autofocus characteristic values, and discrimination analysis is performed on the autofocus performance values ​​and the autofocus characteristic values ​​to obtain discrimination output results of the autofocus performance values ​​and the autofocus characteristic values, the discrimination output results including qualified focus and unqualified focus, if the discrimination output results of the autofocus performance value and the autofocus characteristic value are qualified focus, a mechanical qualified signal is generated, and if the discrimination output results of the autofocus performance value and the autofocus characteristic value are unqualified focus, a mechanical unqualified signal is generated.

[0008] Preferably, the autofocus performance value represents the response time from the moment when the target SWIR6mm lens starts autofocusing to the moment when the focus clarity is equal to a preset focus clarity threshold based on a fixed object distance; the autofocus characteristic value represents the ratio of the number of times the deviation between the actual focus position and the preset focus position of the target SWIR6mm lens at a fixed object distance is greater than the preset threshold and the total number of tests.

[0009] Preferably, the optical performance detection and analysis process is as follows: The optical detection data of the target SWIR6mm lens within the time threshold is obtained, and the optical detection data includes the actual focal length range A, the resolution offset index and the edge distortion rate. The analysis process of the actual focal length range A is as follows: S1: The target SWIR6mm lens is mounted on the set SWIR camera, and aligned with the collimated light beam output by the parallel light tube; S2: The flange distance of the target SWIR6mm lens is adjusted to image the light spot at the center of the sensor, and the actual focal length is recorded; S3: Repeat the test for m groups, where m is a natural number greater than 3, to obtain the maximum and minimum values ​​of the actual focal length in the m groups of tests, and construct the actual focal length range A based on the maximum and minimum values ​​of the actual focal length, and compare and analyze the actual focal length range A to obtain a calibration signal or a non-calibration signal.

[0010] Preferably, the process of obtaining the resolution offset index is as follows: aligning the target SWIR 6 mm lens with a set standard resolution board based on a set fixed distance, using a SWIR camera to capture a characteristic image of the target board, obtaining MTF curves of the center and edge areas of the characteristic image of the target board, and comparing and analyzing the center MTF curve and the edge area MTF curve with the standard center MTF curve and the standard edge area MTF curve, setting the difference value between the center MTF curve and the standard center MTF curve and the difference value between the edge area MTF curve and the standard edge area MTF curve as the center difference and the edge difference, respectively, performing discriminant analysis on the center difference and the edge difference, obtaining the number of abnormal corresponding results of the comparison between the center difference and the edge difference, setting the number of abnormal corresponding results of the comparison between the center difference and the edge difference as the resolution offset index, and performing comparison analysis on the resolution offset index to obtain a conforming signal or a non-conforming signal.

[0011] Preferably, the edge distortion rate is obtained by photographing a grid pattern or a checkerboard calibration plate to obtain an image to be analyzed, analyzing the image to be analyzed to obtain an edge distortion rate, and performing a comparison analysis on the edge distortion rate to obtain a stable signal or a distortion signal; The number of generated calibration signals, compliance signals and stable signals is obtained, and the number of generated calibration signals, compliance signals and stable signals is set as the optical detection coefficient, and the optical detection coefficient is discriminated and processed to obtain an optical compliance signal or an optical deviation signal.

[0012] Preferably, the dual performance detection and analysis process is as follows: a combination of a generated mechanical qualified signal and an optical qualified signal is set as a first identification string, a combination of a generated mechanical qualified signal and an optical offset signal is set as a second identification string, a combination of a generated mechanical unqualified signal and an optical qualified signal is set as a third identification string, and a combination of a generated mechanical unqualified signal and an optical offset signal is set as a fourth identification string; if the first identification string is obtained, a quality output signal is generated; if the first identification string is not obtained, a defect output signal is generated; The total number of defect signals, the second identification string, the third identification string and the fourth identification string obtained during the current batch inspection of the target SWIR 6 mm lens is set as the batch quality deviation rate, and the batch quality deviation rate is compared and analyzed to obtain a batch quality signal or a batch risk signal.

[0013] The beneficial effects of the present invention are as follows: (1) The present invention initially analyzes the basic size of the target SWIR6mm lens and detects the basic size of the target SWIR6mm lens through two points of neural network and laser scanning technology, which helps to improve the detection accuracy on the one hand and sort and manage abnormal target SWIR6mm lenses on the other hand; (2) The present invention further detects the target SWIR6mm lens from the perspectives of mechanical performance and optical performance through information feedback to determine whether the target SWIR6mm lens is qualified, so as to further sort and manage the target SWIR6mm lens. At the same time, the detection output signal is summarized and analyzed to troubleshoot the problems existing in the current batch of the target SWIR6mm lens, so as to improve the production qualification rate of the target SWIR6mm lens. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below in conjunction with the accompanying drawings; Figure 1 It is a flowchart of the system of the present invention; Figure 2 It is a local analysis reference diagram of the first embodiment of the present invention; Figure 3 It is a reference diagram for local analysis of the second embodiment of the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive of other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments;

[0017] Embodiment 1: See also Figures 1 to 3 As shown, the present invention is an industrial detection system for SWIR6mm lens based on neural network, including a lens industrial detection center, a basic detection unit, a scanning analysis unit, a mechanical analysis unit, an optical analysis unit, an evaluation feedback unit and an early warning feedback unit, the lens industrial detection center is connected to the basic detection unit in a one-way communication, the basic detection unit is connected to the scanning analysis unit and the early warning feedback unit in a one-way communication, the scanning analysis unit is connected to the mechanical analysis unit, the optical analysis unit and the early warning feedback unit in a one-way communication, the mechanical analysis unit and the optical analysis unit are connected to the evaluation feedback unit in a one-way communication, and the evaluation feedback unit is connected to the early warning feedback unit in a one-way communication; The lens industrial inspection center is used to collect the appearance feature images of the target SWIR6mm lens, and send the appearance feature images to the basic inspection unit for basic size inspection and analysis, so as to detect the basic size of the target SWIR6mm lens from two angles of neural network and laser scanning technology, which helps to improve the inspection accuracy on the one hand, and helps to sort the target SWIR6mm lens on the other hand. The specific basic size inspection and analysis process is as follows: The detection period of the target SWIR6mm lens is collected, and the detection period of the target SWIR6mm lens is set as the time threshold, and multiple groups of appearance feature images of the target SWIR6mm lens corresponding to qualified or unqualified size are obtained. An appearance feature image set is constructed based on the multiple groups of appearance feature images of the target SWIR6mm lens, and the appearance feature image set is preprocessed. The preprocessing includes denoising and cleaning. At the same time, the preprocessed appearance feature image set is trained through AI deep learning and machine vision technology, and finally a size detection model is obtained; Among them, AI deep learning refers to a subset of machine learning methods based on artificial neural networks (especially deep neural networks), which has learning functions such as representation. Deep learning automatically extracts features from large amounts of data by building and training deep neural network models, and is used to solve complex tasks such as image recognition, speech recognition, and natural language processing; A surface size image of the target SWIR6mm lens is obtained, the surface size image includes a lens flange surface feature image, a threaded interface feature image, etc. The surface size image is input into a size detection model, and output information of the size detection model is obtained, the output information includes qualified surface size and unqualified surface size. The surface size image is input into the size detection model, and the corresponding output information is the number of unqualified surface sizes, which is set as the size measurement index, and the size measurement index is compared and analyzed. If the size measurement index is equal to zero, a normal signal is generated, and if the size measurement index is not equal to zero, an abnormal signal is generated, that is, the basic size of the target SWIR6mm lens is preliminarily detected by the normal signal or the abnormal signal, so as to improve the accuracy of subsequent analysis results.

[0018] The scanning analysis unit is used to compare and analyze the collected surface size information of the target SWIR6 mm lens one by one. The specific one-by-one comparison and analysis process is as follows: The surface dimension information of the target SWIR6mm lens within the time threshold is obtained by laser scanning technology, and the surface dimension information includes the lens flange surface dimension, the thread interface dimension, etc., and the surface dimension information is compared and analyzed with the standard surface dimension information on the design drawing of the target SWIR6mm lens one by one, and the output result of the comparison analysis is obtained, and the output result includes a consistent result and a different result. If the output result of the comparison analysis is a consistent result, a consistent signal is generated, and if the output result of the comparison analysis is a different result, a different signal is generated; Consistent result and difference result: if the surface size information meets the one-to-one correspondence with the standard surface size information on the design drawing of the target SWIR6mm lens, it is determined to be a consistent result; if the surface size information does not meet the one-to-one correspondence with the standard surface size information on the design drawing of the target SWIR6mm lens, it is determined to be a difference result; Interactive output analysis of consistent signals, difference signals, normal signals, and abnormal signals: If a consistent signal and a normal signal are generated, a standard signal is obtained; If a consistent signal and an abnormal signal or a difference signal and a normal signal or a difference signal and an abnormal signal are generated, a defect signal is obtained, and the standard signal or the defect signal is sent to the early warning feedback unit. After receiving the standard signal or the defect signal, the early warning feedback unit immediately makes a preset early warning operation corresponding to the standard signal or the defect signal, that is, the target SWIR6mm lens is classified through the preset early warning operation, so as to sort and manage the target SWIR6mm lenses with defects, and at the same time facilitate in-depth detection of the target SWIR6mm lenses with standard size.

[0019] Embodiment 2: When the standard signal is generated, the mechanical analysis unit is used to perform mechanical quality inspection and analysis on the collected mechanical inspection data of the target SWIR 6 mm lens. The specific mechanical quality inspection and analysis process is as follows: Acquire mechanical inspection data of the target SWIR 6mm lens within the time threshold, the mechanical inspection data including autofocus performance value and autofocus characteristic value, perform discrimination analysis on the autofocus performance value and the autofocus characteristic value, obtain discrimination output results of the autofocus performance value and the autofocus characteristic value, the discrimination output results including focus qualified and focus unqualified, if the discrimination output results of the autofocus performance value and the autofocus characteristic value are focus qualified, generate a mechanical qualified signal, if the discrimination output results of the autofocus performance value and the autofocus characteristic value are focus unqualified, generate a mechanical unqualified signal; The autofocus performance value represents the response time from the moment the target SWIR 6mm lens starts autofocusing to the moment the focus clarity is equal to the preset focus clarity threshold based on the fixed object distance; The autofocus characteristic value indicates the ratio between the number of times the deviation between the actual focus position and the preset focus position of the target SWIR 6 mm lens under a fixed object distance is greater than the preset threshold and the total number of tests; Focus qualified and focus unqualified: If the autofocus performance value is less than or equal to the preset autofocus performance value threshold, and the autofocus characteristic value is less than or equal to the preset autofocus characteristic value threshold, then the focus is judged to be qualified; if the autofocus performance value is greater than the preset autofocus performance value threshold, or the autofocus characteristic value is greater than the preset autofocus characteristic value threshold, then the focus is judged to be unqualified, that is, a preliminary analysis is performed from the perspective of mechanical performance to determine whether the target SWIR6mm lens is normal, and a mechanical qualified signal or a mechanical unqualified signal is generated and sent to the evaluation feedback unit.

[0020] When the standard signal is generated, the optical analysis unit is used to perform optical performance detection and analysis on the collected optical detection data of the target SWIR 6 mm lens. The specific optical performance detection and analysis process is as follows: Obtain optical detection data of the target SWIR 6 mm lens within the time threshold, the optical detection data including the actual focal length range A, the resolution offset index and the edge distortion rate; T1: The analysis process of the actual focal length range A is as follows: S1: Mount the target SWIR 6 mm lens on the set SWIR camera and align it with the collimated beam output by the collimator; S2: Adjust the flange focal distance of the target SWIR6mm lens so that the light spot is imaged at the center of the sensor and record the actual focal length; S3: Repeat the test for m groups, where m is a natural number greater than 3, to obtain the maximum and minimum values ​​of the actual focal lengths in the m groups of tests, and to construct an actual focal length interval A based on the maximum and minimum values ​​of the actual focal lengths; A comparison and analysis is performed on the actual focal length interval A. If the actual focal length interval A is included in the preset focal length interval, a calibration signal is generated. If the actual focal length interval A is not included in the preset focal length interval, a non-calibration signal is generated. T2: The process of obtaining the resolution shift index is as follows: Align the target SWIR6mm lens to the set standard resolution board based on the set fixed distance, use the SWIR camera to shoot the target board characteristic image, obtain the MTF curves of the center and edge areas of the target board characteristic image, and compare and analyze the center MTF curve and the edge area MTF curve with the standard center MTF curve and the standard edge area MTF curve, set the difference value between the center MTF curve and the standard center MTF curve and the difference value between the edge area MTF curve and the standard edge area MTF curve as the center difference and the edge difference, respectively, perform discriminant analysis on the center difference and the edge difference, obtain the comparison result of the center difference and the edge difference as the number of abnormal corresponding ones, and set the comparison result of the center difference and the edge difference as the number of abnormal corresponding ones as the resolution offset index; Comparing and analyzing the resolution shift index, if the resolution shift index is equal to zero, a compliance signal is generated, and if the resolution shift index is not equal to zero, a non-compliance signal is generated; The comparison result is abnormal if: the center difference is greater than the preset center difference threshold or the edge difference is greater than the preset edge difference threshold or the center difference is greater than the preset center difference threshold and the edge difference is greater than the preset edge difference threshold; T3: The process of obtaining the edge distortion rate is as follows: An image to be analyzed is obtained by photographing a grid pattern or a checkerboard calibration plate, an edge distortion rate is obtained by analyzing the image to be analyzed, and the edge distortion rate is compared and analyzed. If the edge distortion rate is less than or equal to a preset edge distortion rate threshold, a stable signal is generated; if the edge distortion rate is greater than the preset edge distortion rate threshold, a distortion signal is generated; For example, when using a grid calibration plate, the geometric features of the pattern must be clear and the repeatability must be high (the error of the grid corner spacing must be ≤0.01mm). The surface of the grid calibration plate must be covered with a high-reflectivity material (such as an aluminum plate) to ensure sufficient imaging contrast in the SWIR band (900-1700nm). The grid calibration plate must cover the entire field of view of the lens and be tilted at multiple angles to cover the edge distortion area to obtain the image to be analyzed. In the embodiment of the present invention, for example, the edge distortion rate is obtained by calculating and processing the image to be analyzed using tools such as OpenCV; T4: The number of generated calibration signals, compliance signals and stable signals is obtained, and the number of generated calibration signals, compliance signals and stable signals is set as the optical detection coefficient, and the optical detection coefficient is judged. If the optical detection coefficient is equal to 3, an optical compliance signal is generated. If the optical detection coefficient is not equal to 3, an optical offset signal is generated, that is, further analysis is performed from the perspective of optical performance to determine whether the target SWIR6mm lens is normal; at the same time, the optical compliance signal or optical offset signal is sent to the evaluation feedback unit.

[0021] The evaluation feedback unit is used to retrieve the mechanical qualified signal, mechanical unqualified signal, optical qualified signal and optical offset signal for dual performance detection and analysis. The specific dual performance detection and analysis process is as follows: The combination of generating a mechanical qualified signal and an optical qualified signal is set as a first identification string, the combination of generating a mechanical qualified signal and an optical offset signal is set as a second identification string, the combination of generating a mechanical unqualified signal and an optical qualified signal is set as a third identification string, and the combination of generating a mechanical unqualified signal and an optical offset signal is set as a fourth identification string; If the first identification string is obtained, a quality output signal is generated; If the first identification string is not obtained, a defect output signal is generated, and the quality output signal or the defect output signal is sent to the early warning feedback unit. After receiving the quality output signal or the defect output signal, the early warning feedback unit immediately displays the preset early warning text corresponding to the quality output signal or the defect output signal, so as to detect whether the target SWIR6mm lens is normal from the perspectives of mechanical performance and optical performance, so as to sort and manage the target SWIR6mm lens; At the same time, the total number of defect signals, the second identification string, the third identification string and the fourth identification string obtained during the current batch detection process of the target SWIR6mm lens is set as the batch quality deviation rate, and the batch quality deviation rate is compared and analyzed. If the batch quality deviation rate is less than the preset batch quality deviation rate threshold, a batch quality signal is generated. If the batch quality deviation rate is greater than or equal to the preset batch quality deviation rate threshold, a batch risk signal is generated, and the batch quality signal or the batch risk signal is sent to the early warning feedback unit. After receiving the batch quality signal or the batch risk signal, the early warning feedback unit immediately displays the preset early warning text corresponding to the batch quality signal or the batch risk signal, that is, the preset early warning text corresponding to the batch quality signal is "batch quality is stable", and the preset early warning text corresponding to the batch risk signal is "batch quality is abnormal", so as to troubleshoot the problems existing in the current batch of the target SWIR6mm lens to improve the production qualification rate of the target SWIR6mm lens; In summary, the present invention preliminarily analyzes from the perspective of the basic size of the target SWIR6mm lens, and detects the basic size of the target SWIR6mm lens through two points of neural network and laser scanning technology, which helps to improve the detection accuracy on the one hand, and helps to sort and manage abnormal target SWIR6mm lenses on the other hand, and further detects the target SWIR6mm lens from the perspectives of mechanical properties and optical properties through information feedback to determine whether the target SWIR6mm lens is qualified, so as to further sort and manage the target SWIR6mm lens, and at the same time, the detection output signal is summarized and analyzed to troubleshoot the problems existing in the current batch of the target SWIR6mm lens, so as to improve the production qualification rate of the target SWIR6mm lens.

[0022] The threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by technicians in this field for each group of sample data; as long as it does not affect the proportional relationship between the parameter and the quantized value.

[0023] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. The industrial inspection system of SWIR6mm lens based on neural network is characterized by: It includes lens industrial testing center, basic testing unit, scanning analysis unit, mechanical analysis unit, optical analysis unit, evaluation feedback unit and early warning feedback unit; The lens industrial inspection center is used to collect the appearance feature image of the target SWIR6 mm lens, and send the appearance feature image to the basic inspection unit for basic size inspection and analysis to obtain a normal signal or an abnormal signal. The scanning analysis unit is used to compare and analyze the collected surface size information of the target SWIR6 mm lens one by one to obtain a consistent signal or a difference signal, and interactively output and analyze the consistent signal, the difference signal, the normal signal and the abnormal signal to obtain a standard signal or a defect signal. When the standard signal is generated, the mechanical analysis unit is used to perform mechanical quality detection and analysis on the collected mechanical detection data to obtain a mechanical qualified signal or a mechanical unqualified signal, and the optical analysis unit is used to perform optical performance detection and analysis on the collected optical detection data to obtain an optical qualified signal or an optical deviation signal; The evaluation feedback unit is used to retrieve the mechanical qualified signal, mechanical unqualified signal, optical qualified signal and optical deviation signal for dual performance detection and analysis to obtain a quality output signal or a defect output signal. At the same time, the batch quality deviation rate is obtained through dual performance detection and analysis, and the batch quality deviation rate is compared and analyzed to obtain a batch quality signal or a batch risk signal.

2. The industrial detection system of SWIR 6 mm lens based on neural network according to claim 1 is characterized in that: The basic size detection and analysis process is as follows: the detection period of the target SWIR6 mm lens is collected, and the detection period of the target SWIR6 mm lens is set as the time threshold, and multiple groups of appearance feature images of the target SWIR6 mm lens corresponding to qualified or unqualified sizes are obtained, and an appearance feature image set is constructed based on the multiple groups of appearance feature images of the target SWIR6 mm lens, and the appearance feature image set is preprocessed, and the preprocessed appearance feature image set is trained by AI deep learning and machine vision technology, and finally a size detection model is obtained; The surface size image of the target SWIR6mm lens is obtained, and the surface size image includes the lens flange surface feature image and the threaded interface feature image. The surface size image is input into the size detection model, and the output information of the size detection model is obtained. The output information includes qualified surface size and unqualified surface size. The output information corresponding to the surface size image input into the size detection model is that the number of unqualified surface sizes is set as the size measurement index, and the size measurement index is compared and analyzed to obtain a normal signal or an abnormal signal.

3. The industrial detection system of SWIR 6 mm lens based on neural network according to claim 2 is characterized in that: The one-by-one comparison and analysis process is as follows: the surface dimension information of the target SWIR6mm lens within the time threshold is obtained through laser scanning technology, the surface dimension information includes the lens flange surface dimension and the threaded interface dimension, and the surface dimension information is compared and analyzed one by one with the standard surface dimension information on the design drawing of the target SWIR6mm lens, and the output result of the comparison and analysis is obtained, and the output result includes a consistent result and a different result. If the output result of the comparison and analysis is a consistent result, a consistent signal is generated, and if the output result of the comparison and analysis is a different result, a different signal is generated.

4. The industrial detection system of SWIR 6 mm lens based on neural network according to claim 3 is characterized in that: The mechanical quality inspection and analysis process is as follows: obtaining mechanical inspection data of the target SWIR6mm lens within a time threshold, the mechanical inspection data including autofocus performance values ​​and autofocus characteristic values, and performing discrimination analysis on the autofocus performance values ​​and the autofocus characteristic values ​​to obtain discrimination output results of the autofocus performance values ​​and the autofocus characteristic values, the discrimination output results including qualified focus and unqualified focus, if the discrimination output results of the autofocus performance values ​​and the autofocus characteristic values ​​are qualified focus, a mechanical qualified signal is generated, if the discrimination output results of the autofocus performance values ​​and the autofocus characteristic values ​​are unqualified focus, a mechanical unqualified signal is generated.

5. The industrial detection system of SWIR 6 mm lens based on neural network according to claim 4 is characterized in that: The autofocus performance value represents the response time from the moment when the target SWIR 6 mm lens starts autofocusing to the moment when the focus clarity is equal to the preset focus clarity threshold value based on the fixed object distance; The autofocus characteristic value represents the ratio between the number of times the deviation between the actual focus position and the preset focus position of the target SWIR 6 mm lens at a fixed object distance is greater than the preset threshold and the total number of tests.

6. The industrial detection system of SWIR 6 mm lens based on neural network according to claim 2 is characterized in that: The optical performance detection and analysis process is as follows: The optical detection data of the target SWIR6mm lens within the time threshold is obtained, and the optical detection data includes the actual focal length range A, the resolution offset index and the edge distortion rate. The analysis process of the actual focal length range A is as follows: S1: The target SWIR6mm lens is mounted on the set SWIR camera, and aligned with the collimated light beam output by the parallel light tube; S2: The flange distance of the target SWIR6mm lens is adjusted to image the light spot at the center of the sensor, and the actual focal length is recorded; S3: Repeat the test for m groups, where m is a natural number greater than 3, to obtain the maximum and minimum values ​​of the actual focal length in the m groups of tests, and construct the actual focal length range A based on the maximum and minimum values ​​of the actual focal length, and compare and analyze the actual focal length range A to obtain a calibration signal or a non-calibration signal.

7. The industrial detection system of SWIR 6 mm lens based on neural network according to claim 6, characterized in that: The process of obtaining the resolution offset index is as follows: aligning the target SWIR 6 mm lens with the set standard resolution board based on a set fixed distance, using a SWIR camera to shoot a characteristic image of the target board, obtaining the MTF curves of the center and edge areas of the characteristic image of the target board, and comparing and analyzing the center MTF curve and the edge area MTF curve with the standard center MTF curve and the standard edge area MTF curve, setting the difference value between the center MTF curve and the standard center MTF curve and the difference value between the edge area MTF curve and the standard edge area MTF curve as the center difference and the edge difference, respectively, and performing a discriminant analysis on the center difference and the edge difference, obtaining the number of abnormal corresponding results of the comparison between the center difference and the edge difference, and setting the number of abnormal corresponding results of the comparison between the center difference and the edge difference as the resolution offset index, and performing a comparison analysis on the resolution offset index to obtain a compliant signal or a non-compliant signal.

8. The industrial detection system of SWIR 6 mm lens based on neural network according to claim 7 is characterized in that: The process of obtaining the edge distortion rate is as follows: obtaining an image to be analyzed by photographing a grid pattern or a checkerboard calibration plate, obtaining an edge distortion rate by analyzing the image to be analyzed, and performing a comparison analysis on the edge distortion rate to obtain a stable signal or a distortion signal; The number of generated calibration signals, compliance signals and stable signals is obtained, and the number of generated calibration signals, compliance signals and stable signals is set as the optical detection coefficient, and the optical detection coefficient is discriminated and processed to obtain an optical compliance signal or an optical deviation signal.

9. The industrial detection system of SWIR 6 mm lens based on neural network according to claim 1, characterized in that: The dual performance detection and analysis process is as follows: a combination of a generated mechanical qualified signal and an optical qualified signal is set as a first identification string, a combination of a generated mechanical qualified signal and an optical offset signal is set as a second identification string, a combination of a generated mechanical unqualified signal and an optical qualified signal is set as a third identification string, and a combination of a generated mechanical unqualified signal and an optical offset signal is set as a fourth identification string. If the first identification string is obtained, a quality output signal is generated; if the first identification string is not obtained, a defect output signal is generated; The total number of defect signals, the second identification string, the third identification string and the fourth identification string obtained during the current batch inspection of the target SWIR 6 mm lens is set as the batch quality deviation rate, and the batch quality deviation rate is compared and analyzed to obtain a batch quality signal or a batch risk signal.

Images