Method for quality control of contact lens

Abstract

Methods for quality control of contact lenses are disclosed herein. An example method includes receiving an input image indicative of a target contact lens. The example method includes, based on an analysis of an input image and using a first learning model, outputting a first foreign material metric and rejection data based on the analysis of the input image. The example method includes outputting an artifact metric using an artifact model and based at least on the rejection data. The artifact model may be implemented based on one or more artifact attributes. The example method includes outputting a second foreign material metric using a second learning model and based at least on the first foreign material metric and the artifact metric. The second learning model may be scale-dependent. The second foreign matter metric indicates at least an acceptance or rejection condition of the target contact lens.

Description

Background Art

[0001] Quality control of contact lenses can rely on inspection and testing techniques. However, improvements are needed. Summary of the invention

[0002] Disclosed herein are methods for quality control of contact lenses and / or contact lens packages. An example method may include receiving an input image indicating a target contact lens. The example method may include analyzing the input image using a first learning model. The first learning model may be scale-independent. The example method may include outputting a first foreign matter metric and rejection data based on the analysis of the input image. The example method may include identifying false rejection data and true rejection data from the rejection data using an artifact model. The artifact model may be implemented based on one or more artifact attributes. The false rejection data may indicate the presence of one or more artifact attributes. The example method may include outputting an artifact metric based on one or more of the false rejection data and the false rejection data. The example method may include outputting a second foreign matter metric using a second learning model and based at least on the first foreign matter metric and the artifact metric. The second learning model may be scale-dependent. The second foreign matter metric may at least indicate an acceptance or rejection condition for the target contact lens.

[0003] The first learning model can be trained and tested on a plurality of images including acceptance or rejection conditions. The artifact model can be trained and tested on at least images including one or more artifact attributes. The one or more artifact attributes can include clear bubbles, folded lenses, or no artifacts, or a combination thereof.

[0004] The second learning model can be trained and tested on at least a plurality of images including acceptance or rejection conditions. The second learning model can be configured to recover false rejections based on the scale of the defect that caused the false rejection. The second learning model can be configured to recover false rejections based on suspected artifacts that are smaller than a predetermined size.

[0005] Methods for quality control of contact lenses are disclosed herein. An example method may include receiving an input image indicative of a target contact lens. The example method may include, based on an analysis of the input image and using a first learning model, outputting a first foreign matter metric and rejection data based on the analysis of the input image. The example method may include outputting the artifact metric using an artifact model and based at least on the rejection data. The artifact model may be implemented based on one or more artifact attributes. The example method may include outputting a second foreign matter metric using a second learning model and based at least on the first foreign matter metric and the artifact metric. The second learning model may be scale-dependent. The second foreign matter metric may at least indicate an acceptance or rejection condition for the target contact lens.

[0006] The first learning model can be trained and tested on a plurality of images including acceptance or rejection conditions. The artifact model can be trained and tested on at least images including one or more artifact attributes. The one or more artifact attributes can include clear bubbles, folded lenses, or no artifacts, or a combination thereof.

[0007] The second learning model can be trained and tested on at least a plurality of images including acceptance or rejection conditions. The second learning model can be configured to recover false rejections based on the scale of the defect that caused the false rejection. The second learning model can be configured to recover false rejections based on suspected artifacts that are smaller than a predetermined size. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The following drawings illustrate various examples discussed in the present disclosure generally by way of example and not by way of limitation. In the drawings:

[0009] Figure 1 An exemplary series model diagram is depicted.

[0010] Figure 2 An exemplary collection diagram is depicted. DETAILED DESCRIPTION

[0011] Disclosed herein are methods for quality control of contact lenses and / or contact lens packages. Lens packages may require quality control to ensure that no foreign matter intrudes into the package. Packages may also require inspection to identify cavities in the package or lens, and edge defects in the lens. Foreign matter may damage the lens or contaminate the lens, both of which are detrimental to meeting customer needs. Packages may be inspected at any stage of the process, including before and after the package is sealed.

[0012] In the following detailed description, reference is made to the accompanying drawings, which form a part of the present disclosure, and in which specific embodiments of the present invention are shown in a graphical manner. It should be understood that other specific implementations and structural changes can be utilized without departing from the scope of the present invention. Without departing from the spirit and scope of the present invention, electrical, mechanical, logical and structural changes can be made to the embodiments. Therefore, the following detailed description should not be understood as limiting, and the scope of the present disclosure is limited only by the appended claims and their equivalents.

[0013] Tandem Models - Tandem models may include an ensemble model in tandem with a traditional deep learning model to help reduce false rejections and maintain high performance. As an illustrative example, a tandem model may be used. First, the ensemble model may include a high-capability deep learning model coupled with an artifact model. The artifact model can be used to recover false rejections due to folded lenses and bubbles. A second deep learning model in tandem with the ensemble model can be used to recover false rejections due to suspicious artifacts smaller than a predetermined size. The first deep learning model can be scale-dependent, while the second deep learning model is scale-independent. The ordering of these models (scale-independent and then scale-dependent) in tandem allows for high performance of the first model (capturing suspicious artifacts regardless of size) and then reducing false rejections (due to size) through the second model. If the order of the models is reversed, the total number of true rejections detected will be reduced due to scale-dependency, resulting in a reduction in the overall tandem model capability. Figure 1 An ordering of ML models forming an example tandem model according to an aspect of the present disclosure is shown.

[0014] Ensemble Models - Ensemble models combine standard deep learning models with artifact models (SVMs). These models can be run in parallel and put into an ensemble model to produce an optimal model with high detection capability and low false rejection rate (see Figure 2 ).

[0015] As an illustrative example, the ensemble was trained with 3018 images (see Table 1 for details).

[0016] Table 1 - Collection of training images

[0017] Ground truth labels Number of images Good image 2320 Bad image 698

[0018] In this case, the purpose of the artifact model is twofold. First, it recovers false rejections from deep learning model 1. Second, it reinforces true rejections from deep learning model 1. The artifact model is trained with 833 images covering 4 main attributes as described in Table 2 below. Other artifact attributes may be used.

[0019] Table 2 - Artifact model training images

[0020] property Number of images Assigned tags Transparent bubbles 89 good fold 144 good No artifacts 500 NA

[0021] Methods for quality control of contact lenses are disclosed herein. These methods may be performed by a computing device. Instructions for performing the method steps may be stored in a memory of the computing device. These method steps may be performed by a processor of the computing device.

[0022] An input image indicative of a target contact lens may be received.

[0023] The input image may be analyzed using a first learning model. The first learning model may be scale-independent. The first learning model may include a deep learning model. The first learning model may be trained and tested on multiple images including acceptance or rejection conditions.

[0024] Based on the analysis of the input image, a first foreign matter metric and rejection data may be output.

[0025] An artifact model from the rejection data may be used to identify false rejection data and true rejection data. The artifact model may be implemented based on one or more artifact attributes. The false rejection data may indicate the presence of one or more artifact attributes. The artifact model may be trained and tested on at least images including the one or more artifact attributes. The one or more artifact attributes may include a transparent bubble, a folded lens, or no artifact, or a combination thereof.

[0026] An artifact metric may be output based on one or more of the false rejection data and the false rejection data.

[0027] A second foreign matter metric may be output using a second learning model and based at least on the first foreign matter metric and the artifact metric. The second learning model may be scale-dependent. The second foreign matter metric may at least indicate an acceptance or rejection condition for a target contact lens. The second learning model may include a deep learning model. The second learning model may be trained and tested on at least a plurality of images including acceptance or rejection conditions. The second learning model may be configured to recover a false rejection based on a scale of a defect that caused the false rejection. The second learning model may be configured to recover a false rejection based on a suspected artifact that is less than a predetermined size.

[0028] Methods for quality control of contact lenses are disclosed herein. These methods may be performed by a computing device. Instructions for performing the method steps may be stored in a memory of the computing device. These method steps may be performed by a processor of the computing device.

[0029] An input image indicative of a target contact lens may be received.

[0030] Based on the analysis of the input image and using the first learning model, a first foreign matter metric and rejection data can be output based on the analysis of the input image. The first learning model can include a deep learning model. The first learning model can be trained and tested on multiple images including acceptance or rejection conditions.

[0031] An artifact model may be used and an artifact metric may be output based at least on the rejection data. The artifact model may be implemented based on one or more artifact attributes. The artifact model may be trained and tested at least on images including the one or more artifact attributes. The one or more artifact attributes may include a clear bubble, a folded lens, or no artifacts, or a combination thereof.

[0032] A second foreign matter metric may be output using a second learning model and based at least on the first foreign matter metric and the artifact metric. The second learning model may be scale-dependent. The second foreign matter metric may at least indicate an acceptance or rejection condition for a target contact lens. The second learning model may include a deep learning model. The second learning model may be trained and tested on at least a plurality of images including acceptance or rejection conditions. The second learning model may be configured to recover a false rejection based on the scale of the defect that caused the false rejection. The second learning model may be configured to recover a false rejection based on a suspected artifact that is less than a predetermined size. Any desired size threshold may be used, such as less than 1000 microns, less than 900 microns, less than 800 microns, less than 700 microns, less than 600 microns, less than 500 microns, less than 400 microns, 300 microns, less than 200 microns, or less than 100 microns in length.

[0033] aspect

[0034] The present disclosure includes at least the following aspects.

[0035] Aspect 1: A method for quality control of contact lenses, the method comprising: receiving an input image indicating a target contact lens; analyzing the input image using a first learning model, wherein the first learning model is scale-independent; outputting a first foreign matter metric and rejection data based on the analysis of the input image; identifying false rejection data and true rejection data from the rejection data using an artifact model, wherein the artifact model is implemented based on one or more artifact attributes, and wherein the false rejection data indicates the presence of the one or more artifact attributes; outputting an artifact metric based on one or more of the false rejection data and the false rejection data; and outputting a second foreign matter metric using a second learning model and based at least on the first foreign matter metric and the artifact metric, wherein the second learning model is scale-dependent, and wherein the second foreign matter metric at least indicates an acceptance or rejection condition of the target contact lens.

[0036] Aspect 2: The method according to Aspect 1, wherein the first learning model comprises a deep learning model.

[0037] Aspect 3: The method according to any one of aspects 1 to 2, wherein the first learning model is trained and tested on a plurality of images including acceptance or rejection conditions.

[0038] Aspect 4: The method according to any one of aspects 1 to 3, wherein the artifact model is trained and tested at least on images including the one or more artifact attributes.

[0039] Aspect 5: The method according to any one of aspects 1 to 4, wherein the one or more artifact attributes include a transparent bubble, a folded lens, or no artifacts, or a combination thereof.

[0040] Aspect 6: The method according to any one of Aspects 1 to 5, wherein the second learning model comprises a deep learning model.

[0041] Aspect 7: The method according to any one of Aspects 1 to 6, wherein the second learning model is trained and tested on at least a plurality of images including acceptance or rejection conditions.

[0042] Aspect 8: The method according to any one of aspects 1 to 7, wherein the second learning model is configured to recover false rejections based on the scale of the defect that caused the rejection.

[0043] Aspect 9: The method according to any one of aspects 1 to 8, wherein the second learning model is configured to recover false rejections based on suspicious artifacts smaller than a predetermined size.

[0044] Aspect 10: A method for quality control of contact lenses, the method comprising: receiving an input image indicating a target contact lens; based on an analysis of the input image and using a first learning model, outputting a first foreign matter metric and rejection data based on the analysis of the input image; using an artifact model and outputting an artifact metric based at least on the rejection data, wherein the artifact model is implemented based on one or more artifact attributes; and using a second learning model and outputting a second foreign matter metric based at least on the first foreign matter metric and the artifact metric, wherein the second learning model is scale-dependent, and wherein the second foreign matter metric at least indicates an acceptance or rejection condition for the target contact lens.

[0045] Aspect 11: The method according to Aspect 10, wherein the first learning model comprises a deep learning model.

[0046] Aspect 12: A method according to any one of aspects 10 to 11, wherein the first learning model is trained and tested on a plurality of images comprising acceptance or rejection conditions.

[0047] Aspect 13: The method according to any one of aspects 10 to 12, wherein the artifact model is trained and tested at least on images including the one or more artifact attributes.

[0048] Aspect 14: The method according to any one of aspects 10 to 13, wherein the one or more artifact attributes include a transparent bubble, a folded lens, or no artifact, or a combination thereof.

[0049] Aspect 15: The method according to any one of Aspects 10 to 14, wherein the second learning model comprises a deep learning model.

[0050] Aspect 16: A method according to any one of Aspects 10 to 15, wherein the second learning model is trained and tested on at least a plurality of images including acceptance or rejection conditions.

[0051] Aspect 17: The method according to any one of aspects 10 to 16, wherein the second learning model is configured to recover false rejections based on the scale of the defect that caused the rejection.

[0052] Aspect 18: The method according to any one of aspects 10 to 17, wherein the second learning model is configured to recover false rejections based on suspicious artifacts smaller than a predetermined size.

[0053] Aspect 19: A method according to any one of Aspects 1 to 18, wherein the second learning model is configured to recover false rejections based on suspicious artifacts having a length of less than 1000 microns, less than 900 microns, less than 800 microns, less than 700 microns, less than 600 microns, less than 500 microns, less than 400 microns, 300 microns, less than 200 microns or less than 100 microns.

[0054] In the system or method, once an image has been acquired, image preprocessing can be performed before the image is analyzed with the model. Image preprocessing can include identifying specific areas of interest, enhancing contrast, adding or subtracting other images taken, etc.

[0055] Convolutional neural networks (CNNs) may be designed or modified and implemented to identify packages to be rejected. Packages may be sealed or unsealed and may contain lenses or may be empty. In addition to containing lenses or not containing lenses, packages may also contain liquids or not contain liquids. There are a variety of platforms for creating CNNs, and those skilled in the art will understand these platforms, including Tensorflow (Google), Caffe (UC Berkeley), CNTK (Microsoft), Theano (LISA Labs), Keras (Google), Matlab (Mathworks), Octave & Phython. Existing modules are modified to slice the original image into smaller sizes, with only the size of the region of interest (ROI) appropriately set to feed into the CNN. These images are then defined by a region of interest (ROI) of, for example, 299×299 pixels (monochrome), which is defined by the package itself, so that each ROI fed into the CNN has an appropriate number of pixels for processing. An exemplary CNN such as Inception V3 may include a total of 48 layers mixed with convolutional layers, maximum pooling layers, and fully connected layers. In the later stages of the network, ReLU (Rectified Linear Units) can be used and a large number of nodes can be dropped in a random manner to avoid overfitting. For example, a filter of size 3×3 can be used with a stride of 1, 2, or a larger number of strides. The layers can use margins around the edges and corners of the image to enhance the processing of the image. Similarly, the layers of the network (e.g., convolutional layers, max pooling layers, softmax, dropout, fully connected layers, etc.) can be changed to achieve better results in identifying packages with some defects while not incorrectly identifying packages without defects as defective or containing FM.

[0056] Many operating systems, including Linux, and An operating system that is capable of running many tasks simultaneously is called a multitasking operating system. Multitasking is the ability of an operating system to execute more than one executable task at the same time. Each executable task runs with its own address space, which means that the executable tasks cannot share their memory. Therefore, it is not possible for any program to damage the execution of any other program that is running on the system. However, programs cannot exchange any information through the operating system (or by reading files stored on the file system).

[0057] Multi-process computing is similar to multi-tasking computing in that the terms task and process are often used interchangeably, although some operating systems distinguish between the two. The present invention may be or include a system, method, and / or computer program product at any possible level of technical detail integration. The computer program product may include a computer-readable storage medium (or multiple media) having computer-readable program instructions thereon, which are used to cause a processor to perform various aspects of the present invention. A computer-readable storage medium may be a tangible device that can retain and store instructions for use by an instruction execution device.

[0058] The computer readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of computer readable storage media includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as a punch card or a raised structure in a groove with instructions recorded thereon, and any suitable combination of the foregoing.

[0059] As used herein, computer-readable storage media should not be understood as transient signals themselves, such as radio waves or other free propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses passing through fiber-optic cables), or electrical signals transmitted by wires. Computer-readable program instructions as described herein can be downloaded from computer-readable storage media to corresponding computing / processing devices, or downloaded to external computers or external storage devices via a network (e.g., the Internet, a local area network, a wide area network, and / or a wireless network). A network may include copper transmission cables, transmission optical fibers, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer-readable program instructions from a network and forwards computer-readable program instructions to be stored in a computer-readable storage medium in a corresponding computing / processing device.

[0060] The computer-readable program instructions for performing the operations of the present invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, configuration data for an integrated circuit, or source code or object code written in any combination of one or more programming languages ​​(including object-oriented programming languages, such as Smalltalk, C++, etc.) and procedural programming languages ​​(such as the "C" programming language or similar programming languages). The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., by using the Internet of an Internet service provider).

[0061] In some embodiments, to perform various aspects of the present invention, electronic circuits including, for example, programmable logic circuits, field programmable gate arrays (FPGAs), or programmable logic arrays (PLAs) may execute computer-readable program instructions to personalize the electronic circuits by utilizing state information of the computer-readable program instructions.

[0062] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each box in the flowchart illustration and / or block diagram and the combination of boxes in the flowchart illustration and / or block diagram can be implemented by computer-readable program instructions. These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to generate a mechanism so that instructions executed by the processor of the computer or other programmable data processing device create instructions for implementing the flowchart and / or block diagram. Figure 1 A tool that specifies functions / actions in one or more boxes.

[0063] These computer-readable program instructions may also be stored in a computer-readable storage medium, which may guide a computer, a programmable data processing device, and / or other apparatus to work in a specific manner, so that the computer-readable storage medium having instructions stored therein includes an implementation flowchart and / or block diagram. Figure 1The computer-readable program instructions may also be loaded onto a computer, other programmable data processing device, or other apparatus, so that a series of operation steps are executed on the computer, other programmable device, or other apparatus to produce a computer-implemented process, so that the instructions executed on the computer, other programmable device, or other apparatus implement the flowchart and / or block diagram. Figure 1 The functions / actions specified in one or more boxes. The flowcharts and block diagrams in the accompanying drawings illustrate the possible specific implementation architectures, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram may represent a module, fragment or portion of an instruction, which includes one or more executable instructions for implementing a specified logical function.

[0064] In some alternative implementations, the functions indicated in the frame may not occur in the order indicated in the accompanying drawings. For example, depending on the functions involved, the two frames shown in succession may actually be performed substantially simultaneously, or in reverse order. It should also be noted that each frame in the block diagram and / or the flow chart diagram and the combination of frames in the block diagram and / or the flow chart diagram can be implemented by a dedicated hardware-based system that performs a specified function or action or by a combination of dedicated hardware and computer instructions. Although specific embodiments of the present invention have been described, it will be appreciated by those skilled in the art that there are other embodiments that are equivalent to the described embodiments. Therefore, it should be understood that the present invention is not limited by the specific exemplified embodiments, but is only limited by the scope of the appended claims.

[0065] As can be seen from the above description, the present invention provides systems, computer program products and methods for effectively executing the technology. Unless explicitly stated otherwise, references to singular elements in the claims are not intended to mean "one and only one", but "one or more". All structural and functional equivalents of the elements of the above exemplary embodiments currently known or later known to those of ordinary skill in the art are intended to be covered by the claims of the present invention. Unless explicitly stated using the phrase "device for..." or "step for...", the claim elements herein should not be interpreted according to the provisions of the sixth paragraph of Section 112 of 35 U.S.C.

[0066] Although the foregoing written description of the present invention enables a person of ordinary skill to make and use what is currently considered to be the best mode of the present invention, a person of ordinary skill will understand and appreciate the existence of alternatives, adaptations, variations, combinations and equivalents of the specific embodiments, methods and examples herein. It will be understood by those skilled in the art that the present disclosure is merely exemplary and that various modifications may be made within the scope of the present invention. In addition, although specific features of the teachings have been disclosed with respect to only one of several specific implementations, such features may be combined with one or more other features of other specific implementations, as may be desired and advantageous for any given or specific function. In addition, with respect to the terms "containing", "including", "having", "with" or variations thereof used in the specific embodiments and claims, such terms are intended to be inclusive in a manner similar to the term "comprising".

[0067] Other embodiments of the present teachings will be apparent to those skilled in the art in view of the description and practice of the present teachings disclosed herein. Therefore, the present invention should not be limited to the embodiments, methods and examples described, but rather encompasses all embodiments and methods within the scope and spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments shown herein, but is only limited by the following claims.

Claims

1. A method for quality control of contact lenses, the method comprising: receiving an input image indicative of a target contact lens; Analyzing the input image using a first learned model, wherein the first learned model is scale-independent; outputting a first foreign matter metric and rejection data based on said analyzing of said input image; identifying false rejection data and true rejection data from the rejection data using an artifact model, wherein the artifact model is implemented based on one or more artifact attributes, and wherein the false rejection data indicates the presence of the one or more artifact attributes; outputting an artifact metric based on one or more of the false rejection data and the false rejection data; as well as outputting a second foreign matter metric using a second learned model and based at least on the first foreign matter metric and the artifact metric, wherein the second learned model is scale-dependent, and Wherein the second foreign matter metric is indicative of at least an acceptance or rejection condition of the target contact lens.

2. The method of claim 1, wherein the first learning model comprises a deep learning model.

3. The method of claim 1, wherein the first learning model is trained and tested on a plurality of images including acceptance or rejection conditions. 4 . The method of claim 1 , wherein the artifact model is trained and tested on at least images including the one or more artifact attributes.

5. The method of claim 1, wherein the one or more artifact attributes include a clear bubble, a folded lens, or no artifact, or a combination thereof. The method according to claim 1 , wherein the second learning model comprises a deep learning model.

7. The method of claim 1, wherein the second learning model is trained and tested on at least a plurality of images including acceptance or rejection conditions.

8. The method of claim 1, wherein the second learning model is configured to recover false rejections based on a dimension of a defect that caused the rejection.

9. The method of claim 1, wherein the second learning model is configured to recover false rejections based on suspicious artifacts smaller than a predetermined size.

10. A method for quality control of contact lenses, the method comprising: receiving an input image indicative of a target contact lens; outputting a first foreign matter metric and rejection data based on the analysis of the input image and using a first learned model, wherein the first learned model is scale-independent; outputting an artifact metric using an artifact model and based at least on the rejection data, wherein the artifact model is implemented based on one or more artifact attributes; and outputting a second foreign matter metric using a second learned model and based at least on the first foreign matter metric and the artifact metric, wherein the second learned model is scale-dependent, and Wherein the second foreign matter metric is indicative of at least an acceptance or rejection condition of the target contact lens.

11. The method of claim 10, wherein the first learning model is trained and tested on a plurality of images comprising acceptance or rejection conditions.

12. The method of claim 10, wherein the artifact model is trained and tested on at least images including the one or more artifact attributes.

13. The method of claim 10, wherein the one or more artifact attributes include a clear bubble, a folded lens, or no artifact, or a combination thereof.

14. The method of claim 10, wherein the second learning model is trained and tested on at least a plurality of images including acceptance or rejection conditions.

15. The method of claim 10, wherein the second learning model is configured to recover false rejections based on a dimension of a defect that caused the rejection.

16. The method of claim 10, wherein the second learning model is configured to recover false rejections based on suspicious artifacts smaller than a predetermined size.

17. A method for quality control of contact lenses, the method comprising: receiving an input image indicative of a target contact lens; outputting a first foreign matter metric and rejection data based on the analysis of the input image and using a first model, wherein the first learned model is scale-independent; outputting an artifact metric using an artifact model and based at least on the rejection data, wherein the artifact model is implemented based on one or more artifact attributes; and outputting a second foreign matter metric using a second model and based at least on the first foreign matter metric and the artifact metric, where the second model is scale-dependent, and Wherein the second foreign matter metric is indicative of at least an acceptance or rejection condition of the target contact lens.

18. The method of claim 17, wherein the first learning model is trained and tested on a plurality of images comprising acceptance or rejection conditions.

19. The method of claim 17, wherein the artifact model is trained and tested on at least images including the one or more artifact attributes.

20. The method of claim 17, wherein the one or more artifact attributes include a clear bubble, a folded lens, or no artifact, or a combination thereof.

21. The method of claim 17, wherein the second learning model is trained and tested on at least a plurality of images comprising acceptance or rejection conditions.

22. The method of claim 17, wherein the second learning model is configured to recover false rejections based on a dimension of a defect that caused the rejection.

23. The method of claim 17, wherein the second learning model is configured to recover false rejections based on suspicious artifacts smaller than a predetermined size.

Images