PROCEDURE AND SYSTEM FOR QUALITY CONTROL OF PROCESSED FOOD PRODUCTS

Abstract

In a first aspect, the invention concerns a method for quality control of processed food in a production line using a camera, comprising steps such as obtaining a digital baseline model of a food product, generating a first dataset with variations on this baseline model, and a second dataset with digital models of defects. These are combined into a third dataset of food products with defects. Furthermore, the method includes simulating a production environment in which the food models with defects are placed, generating two-dimensional images of the food products in this production environment, training an AI model with these images to recognize defects, and applying the AI ​​model to identify defective food products in a production line.In a second aspect, the invention comprises a quality control device with a camera, an image processing computer, and a graphical user interface. The computer contains a trained AI model capable of identifying defects in the production line. The invention offers an efficient, cost-effective method for training and deploying an AI model for defect detection, ensuring accurate quality control without affecting production speed. Thanks to automatic data analysis and flexibility regarding product changes, the AI ​​model can be quickly adapted to new requirements in the production environment, thereby guaranteeing food safety and consistency.

Description

BE2024 / 5836 2 segmentation takes place via a 3D neural network trained with manual annotations. Subsequently, a robot command is calculated for handling the object, which is executed by a robot element. Traditional methods for annotating data take a lot of time and are therefore unsuitable for annotating data for the quality control of processed food. The current invention aims to find a solution to this problem. SUMMARY OF THE INVENTION 10 In a first aspect, the invention concerns a method for quality control of processed food in a production line with a camera,the method comprising the steps of: obtaining a digital base model of a food product; obtaining a first dataset comprising digital models of food products by introducing variations on the digital base model; obtaining a second dataset comprising various generated digital models of the defects; obtaining a third dataset comprising digital models of food products with the defects by combining the first and the second dataset; simulating a production environment,and combining the simulated production environment with at least one digital model of a food product; generating two-dimensional images of one or more digital models of food products with the defects in a production environment; training an AI model with the two-dimensional images to recognize food products with the defects; obtaining image material of the processed food in the production line; identifying food products with the defects by analyzing the image material with the trained AI model; where the first dataset is obtained by, with the aid of algorithms,varying the dimensions of the digital basic model and applying surface characteristics to the digital basic model; and whereby different types of defects are generated during the acquisition of the second data set; and whereby during the acquisition of the data set the type of defect and the location of the defect on the digital model are varied per digital model. In a second aspect, the invention concerns a device for quality control of processed food in a production line with a camera, the device comprising: at least one camera aimed at the processed food in the production line; a computer35 capable of receiving images from at least one camera, and of identifying the defective food products based on the received images; a graphical user interface capable of displaying data based on the 2024 / 5836 BE2024 / 5836 3 identified defective food products; and where the computer includes a memory with an AI model trained with generated labeled data of food products without the defects in a simulated production environment,and where the computer is suitable for identifying defective food products with the aid of the AI ​​model.5 Quality control of processed food products in a production line is essential to ensure food safety, consistency, and customer satisfaction. During processing, products can acquire damage, contamination, or deviations in texture, shape, or color. Strict controls help to identify and remove defective products in a timely manner, which minimizes health risks for consumers and limits waste. Because large volumes of food products must move quickly through the production line in the food production industry, manual quality control is practically unfeasible and would significantly slow down production speed. By using an AI model to analyze imagery of the products, defects can be recognized faster and more accurately. The present invention offers an efficient way to train and deploy an AI model for identifying defective food products,by automatically generating annotated data of the defective products. This saves costs and improves quality control. Moreover, this method makes it possible to create a greater variety and quantity of defect models, so that the AI ​​model learns to recognize a broader range of defects and the accuracy of quality control increases. The invention also offers flexibility, because new datasets can be easily generated when the production line or the types of food products change, allowing the AI ​​model to be quickly and effectively adapted to new requirements in the production environment. DETAILED DESCRIPTION Unless otherwise defined, all terms used in the description of the invention, including technical and scientific terms, have the meaning as they are generally understood by the professional in the technical field of the invention. For a better assessment of the description of the invention, the following terms are explicitly explained. 35 “A”,“de” and “het” in this document refer to both the singular and the plural unless the context clearly implies otherwise. For example, “a segment” means one or more than a segment. 2024 / 5836 BE2024 / 5836 4 The terms “comprehensive”, “comprehensive”, “consist of”, “consisting of”, “provided with”, “contain”, “containing”, “encompass”, “encompassing”, “content”, “containing” are synonyms and are inclusive or open terms that indicate the presence of what follows, and that do not exclude or prevent the presence of other components, characteristics, elements, members, steps,known from or described in the standard5 technique. The 'digital base model' refers to a first digital representation of a food product. The term 'digital models' refers to digital representations of food products obtained by applying variations to the digital base model. In a first aspect, the invention concerns a method for quality control of processed food in a production line with a camera according to conclusion 1. Quality control of processed food products in a production line is essential for ensuring food safety, consistency and customer satisfaction. During processing, products may become damaged, contaminated or show deviations in texture, shape or color. Strict quality controls make it possible to detect and remove defective or deviating products early, which helps to minimize health risks for consumers and reduce waste. In the food production industry, quality control poses a major challenge,because25 food products must be transported in high volumes and at high speed to achieve the required efficiency. This makes it practically impossible to manually check quality, as this would significantly slow down the production speed. By using an AI model to analyze imagery of food products, defects can be identified faster and more accurately30. To ensure that the AI ​​model correctly recognizes the defective products, it must first be trained with annotated data of the defective food products. These annotations help the AI ​​model learn what should be considered a defect. 35 The method according to conclusion 1 accelerates the training of the AI ​​model by generating annotated data of the defective food products, instead of manually annotating data of the defective food products. Because manual 2024 / 5836 BE2024 / 5836 5 annotation is redundant, a lot of time and personnel costs are also saved,and many human errors are avoided. In addition, generating models of the defective food products provides more freedom, allowing for a greater variety of models of the defective food products to be generated, enabling the AI ​​model trained with the aforementioned digital models to recognize a wider variety of defects. This freedom also offers the possibility to generate a larger quantity of digital models per defect, which improves the accuracy of the AI ​​model, and thereby quality control. Furthermore, this method offers the possibility to adapt the AI ​​model flexibly and efficiently to changes in the production line, which is crucial given the dynamic nature of modern food production. Production lines are regularly optimized or modified to process new types of food products. In this context, food is often seasonal,causing production lines to frequently switch food products. Consequently, only limited time is available for adjusting quality control. However, with the current method, the AI ​​model can be easily and quickly adapted by generating new annotated datasets for the adjusted food products. Generating digital models of food products and digital models of defects separately, and subsequently combining them, offers significant advantages for the effectiveness and flexibility of the dataset. The separated approach creates separate control over the models of defects and products, meaning that when changes occur, only one of the two needs to be adjusted, saving duplication of work. Moreover, this separation makes it possible to generate a larger dataset, as defects and products can be varied independently of each other. This results in a broader and more realistic range of training data for the AI ​​model. Additionally, it offers the possibility to analyze specific defects and adapt them for different product types.which leads to an accurate and more targeted detection system, tailored to specific inspection criteria and food products in the production line. Furthermore, it can be more easily verified in what proportions certain defects occur on a single digital food product model or across a collection of digital models of food products, whereby the dataset reflects the actual defect frequencies in the production environment. This results in a better trained AI model that effectively responds to realistic defect patterns, which increases the reliability and precision of quality control. 2024 / 5836 BE2024 / 5836 6 In an implementation form, the AI ​​model is partially trained with manually annotated data of actual food products. Preferably, the training data for training the AI ​​model consists of a maximum of 50% manually annotated data, and even more preferably a maximum of 35%,and almost even more preference maximum 20%. 5 In a single implementation form, both the digital models of food products and the production environment are simulated in three dimensions. This offers the advantage that various two-dimensional images can be generated from a single 3D setup, with variations in both product and environmental perspectives. The lighting, shadow formation, and relative positions of objects in the production environment can be easily adjusted without having to create new models or a new environment each time. This approach saves time and increases efficiency, because a wide range of realistic inspection scenarios can be simulated. As a result, the AI ​​model is robustly trained to consistently recognize defects, regardless of variations in the production environment. Through this flexible 3D simulation, the model is better prepared for the actual variations in the production line,which significantly increases the accuracy and reliability of quality control. By adding a simulated production environment to the digital models of food products, a realistic recontext is created, allowing the AI-20 model to learn to recognize not only individual products but also products within a production environment. This ensures that the AI ​​model is trained to identify objects in a complete, complex context without being influenced by randomness in the background. These background variations are varied in the simulated environment, so that the AI ​​model learns to focus on the product itself,25 which prevents the model from drawing incorrect conclusions. By varying the dimensions of the digital models of the food products, a wide range of shapes and proportions can be generated, whereby the first dataset reflects the natural variation in food products. By varying the dimensions before applying the surface characteristics,preserving surface characteristics such as textures and color their proportions, which prevents them from deforming unnaturally with changes in size. This ensures that the digital models are more realistic, which leads to better training of the AI ​​model, resulting in better quality control.35 In an implementation form, during the step of obtaining the first dataset, boundary values ​​are determined based on observed variations of dimensions 2024 / 5836 BE2024 / 5836 7 of the processed food, and where the algorithms are suitable to counteract variations in dimensions based on the determined boundary values. The boundary values ​​prevent unrealistic digital models from being generated, and through the use of algorithms a wide variety of dimensions can be efficiently generated.5 In an implementation form, a Monte Carlo algorithm is applied,where random values ​​within the established boundaries are generated to create variations in dimensions. This algorithm offers a diverse dataset without unrealistic deviations occurring.10 In one implementation form, a Gaussian-noise algorithm is applied, which adds small, natural variations around and mean to the dimensions of the model. This simulates subtle differences that typically occur in production.15 In a subsequent implementation form, a parametric model is applied, where parameters such as height, width, and depth are varied within the boundaries according to predefined rules. This approach offers precise control over specific dimensions,while natural variability is preserved. By using this algorithm, a wide set of realistic models can be generated quickly and efficiently that closely match the actual variations in the processed food product. In one implementation form, various lighting effects and different textures are applied to the digital base model during the step of applying surface characteristics. Lighting effects play a crucial role because light intensities and directions can vary in the actual production environment, which influences the way the surface of food products looks. Variations in lighting can change the visibility of textures, causing surfaces to appear shinier or duller depending on the angle of light. By simulating these effects in the dataset, the AI ​​model learns to recognize defects despite changes in the visual presentation of textures under varying lighting conditions. This significantly improves the accuracy and reliability of the model.because it is prepared for realistic inspection conditions. 35 In an execution form, the incidence of light on the digital models of food products is simulated from different angles, so that unique shadow patterns are created that bring out the texture and shape details more clearly. By applying lighting from various 2024 / 5836 BE2024 / 5836 8 directions, varying shadow zones are created that accentuate both height differences and surface features. This is especially important to make small, but crucial details visible, such as subtle irregularities, ridges, or small indentations, which might remain hidden with uniform, diffuse lighting. As a result, the AI ​​model is better trained to recognize defects and when different5 shadow effects occur on the food products. In one execution form, variations in color and temperature are simulated on the digital models of food products, from warm to cool light,to allow the AI ​​model to adapt to possible color shifts in a production environment. This is important10 because factories have various light sources, such as ambient lighting, signal lighting on machines, and specific work lighting, which can emit different color temperatures. Additionally, windows at the production line can let in natural light variations, ranging from cool morning light to warm evening light, depending on the time of day and weather conditions. By applying these15 light color variations to the digital models, the AI ​​model becomes better prepared for environmental changes, allowing it to continue recognizing defects reliably, regardless of the color of the light. This increases the consistency and accuracy of the inspection under varying production conditions. 20 In a specific form, both smooth and rough textures are applied to the digital models of food products to simulate the surface properties of these products. This allows the AI ​​model to learn to recognize defects,even when the structure of the product surface varies. 25 In another implementation form, the reflectivity of the textures on the digital models of food products is adjusted so that surfaces can appear both matte and glossy. This prevents the AI ​​model from incorrectly interpreting points of shine as defects or overlooking actual defects due to reflection effects on the surface. 30 In an implementation form, color variations are added to the textures of the digital models of food products to simulate natural color differences. This helps the AI ​​model to distinguish between normal color variations and real discolorations that may indicate a defect.35 In one implementation form, ray tracing is used to simulate lighting effects on the digital models of the food products, whereby, among other things, realistic light rays from different angles are simulated. This is beneficial for accurately rendering shadow formations and reflections,so that the AI ​​model learns better how light and shadow influence the visibility of textures and surface details. 5 In another implementation form, global illumination is applied to simulate lighting effects on the digital models of the food products, whereby light gradients and ambient light, among other things, are mimicked. This is beneficial for creating subtle shadows and color shifts, resulting in more natural lighting comparable to the conditions in a real production environment. In another implementation form, Phong shading is used to simulate lighting effects on the digital models of the food products, to realistically render the reflection and gloss of surfaces. This helps train the AI ​​model to distinguish highlights of, for example, smooth surfaces from real defects and prevents misinterpretations of light reflections. In a sequential execution form, ambient inclusion is used to simulate lighting effects on the digital models of the food products,to enhance shadow zones in corners and tight spaces. This is beneficial for rendering depth and texture differences, which helps the AI ​​model to more accurately recognize small defects in hard-to-see parts of the product. In a subsequent execution form, ambient inclusion is applied to enhance shadow zones in corners and tight spaces, which accentuates the sense of depth and texture differences on the models. This facilitates the AI ​​model's detection of small defects in awkward corners or surfaces that might otherwise go unnoticed, which improves overall defect detection. In an execution form, bump mapping is used to simulate a texture on the surface of the digital models of food products, where small height differences are mimicked without changing the underlying geometry. This is beneficial for creating a realistic relief on surfaces,so that the AI ​​model learns to better distinguish between natural textures and possible defects. 2024 / 5836 BE2024 / 5836 10 In another implementation form, displacement mapping is applied to simulate a texture on the surface of the digital models of food products, where the surface is actually deformed based on texture data. This is beneficial for accurately rendering coarse textures, such as ridges or deep indentations, which helps train the AI ​​model to recognize defects on uneven surfaces. In another implementation form, specular mapping is used to simulate a texture on the surface of the digital models of food products, where reflectivity and gloss are adjusted to represent variations in glossy and matte areas. This is beneficial for simulating surfaces that are partially reflective,so that the AI ​​models to distinguish light reflection from actual defects. In a subsequent execution form, normal mapping is used to simulate a texture on the surface of the digital models of food products, by adjusting normal values ​​to mimic small details and reliefs without changing the geometry. This is especially useful for simulating fine surface details,which helps the AI ​​model to recognize subtle texture differences that may indicate defects.20 In one implementation form, the method includes the step of adding details to the digital models using a Generative AI model. In one implementation form, the method includes the step of adding details to the simulated production environment using a Generative AI model. In one implementation form, the images of one or more digital models of food products with the defects are generated in a production environment using a Generative AI model.30 In one implementation form, the Generative AI model concerns Generative Adversarial Networks (GANs). In another implementation form, the Generative AI model concerns a diffusion model.35 This offers the advantage that the reality of both the digital models and the simulated production environment can be magnified by continuously adjusting them,preferably based on imagery from the actual production environment. This allows subtle lighting effects and variations on food products to be simulated more accurately, enabling the AI ​​model to handle realistic lighting influences. The model can also better recognize the finer details of surface characteristics such as texture and gloss, because these are aligned with actual product properties. In addition, realistic changes in the production environment itself can be added, such as the varying presence of objects or personnel, making the AI ​​model more robust in the face of variations in the background. Furthermore, this approach offers the possibility to align the digital models so that they correspond to the exact perspective of one or more cameras used for actual defect detection in the production line. This ensures that the AI ​​model is trained realistically,which significantly improves accuracy and reliability in quality control. In one execution form, during the generation of two-dimensional images, the lighting is varied based on possible lighting conditions on the production line. This ensures not only realistic images, but also images that correspond more accurately to the actual lighting in the production environment. This trains the AI ​​model on specific light variations it actually encounters, such as changes in light intensity, color temperature, or angle of incidence. This increases the accuracy of the model. In one execution form, digital models of food products are simulated in a single production environment without defects or overlaps. Since food products are often transported in bulk through the production line, they frequently overlap and often lie in each other's shadows, causing defects to sometimes be only partially visible or even partially hidden. By incorporating this overlapping and shadow formation into the simulation,the AI ​​model is trained to recognize defects under conditions where they are not fully visible or are visually disturbed by other products. This significantly improves the accuracy of the AI ​​model.30 In one implementation form, motion blur is applied to the two-dimensional images of the digital models. During transport on the production line, food products are often subject to vibrations and movements, which can result in a certain degree of motion blur in the camera footage. By incorporating this blur into the simulation, the AI ​​model is trained to recognize defects under realistic conditions where the products are not completely stationary. This improves the accuracy of the model, 2024 / 5836 BE2024 / 5836 12 because it learns to detect defects despite slight blurring caused by movement,resulting in a more reliable inspection process that better aligns with the actual conditions in the production line. In one implementation form, the motion blur is based on actual footage of the food products in the production line. The degree and nature of this blur are strongly dependent on the specific characteristics of the product and the conditions on the production line. Different products exhibit different types of movements and vibrations during transport, depending on their shape, weight, and method of placement. In addition, factors such as the speed of the production lines and the positioning of the camera (for example, the distance to the production lines and the angle at which filming takes place) can have a major influence on the degree of blur in the footage. It is therefore difficult to accurately simulate realistic motion blur without using actual video material. By adjusting the blur to these specific conditions,15 closely approximates real-world situations and teaches the AI ​​model to detect defects under the real visual challenges of the production environment. In an execution form, the two-dimensional images are generated based on one or more videos that measure realistic dynamic interactions between digital models of food products without defects, where these interactions are based on video footage of the actual production line. By adding dynamic interactions between digital models of food products, the occurrence of unrealistic configurations is prevented, such as models that partially or completely occupy the same space. This ensures that the relative positions of products on the production line are more realistic and correspond better to real-world conditions. An additional benefit is that these interactions automatically generate a wide range of different configurations,resulting in a faster and simpler creation of a greater variety of training data. By basing the interaction on actual video footage of the production line, realism is further enhanced, because the complexities and variation of movements and positions between food products are so difficult to estimate without this reference. This ensures that the AI ​​model becomes even more accurate and robust in defect recognition, because it is trained on realistic and varied situations that accurately reflect the actual behavior of the products on the line. In a further implementation form, physical interactions such as gravity and collision dynamics are added to the simulation, allowing food products to realistically stack, shift, and fall depending on their position on the production line. This offers the advantage that the AI ​​model is trained on defect detection in situations where products physically influence each other,such as when stacking or tilting products due to their weight and shape. 5 In another further implementation form, friction is added to the digital models so that products realistically slide or decelerate upon contact with other objects. This ensures that the AI ​​model is exposed to realistic deceleration and acceleration patterns, which often occur when food products touch each other or move across surfaces. 10 In another further implementation form, partial elasticity of products is simulated so that the degree of deflection or deformation during collisions can be mimicked. This helps the AI ​​model to recognize defects, even when products change shape due to pressure or collisions, which often happens in bulk transport. 15 In one implementation form, the perspective is varied during the generation of two-dimensional images, which offers several benefits for quality control. Because production lines are constantly evolving and are often modified or renewed,The cameras for defect detection are regularly moved or placed in a different configuration. By training the AI ​​model to recognize defects from different perspectives, the model does not need to be retrained when the camera positions change, which saves time and costs. A further advantage is that the defects look different from every perspective due to the changing configuration of products, light incidence, and shadow, whereby the AI ​​model is trained on a wider variety of data. This increases accuracy, as it models to recognize defects in diverse visual conditions. Moreover, this approach makes it possible to deploy multiple cameras to detect defective food products without only one camera functioning optimally. After all, the AI ​​model is prepared for various viewpoints, which offers a great deal of freedom when scaling up quality control. In an execution form, the digital basic model of the food product is formed by means of at least 100 photos,taken from multiple perspectives of the product. This offers the advantage of immediately obtaining a realistic representation of the food product. Due to the complexities and irregular characteristics of food products, it is often difficult to design all details manually; however, by basing the model on photos, the generation of digital models begins directly from a true-to-life base model. Moreover, the use of photos from different angles ensures that the shapes and texture of the product are captured more fully, forming a detailed and accurate basis for further model variations. 5 In a production form, the digital base model of the food product is formed using laser scanning. Laser scanning offers the advantage of having extremely high accuracy, allowing even the smallest details of the food product to be captured. This makes it possible to digitize complex shapes and surfaces, such as those of natural products, in great detail.10 which is crucial for reliable quality control and defect detection. In a second aspect, the invention concerns a device in accordance with conclusion 13. This device offers the major advantage that it requires only one camera to identify defective food products, without the need for major modifications to the 15 production line. By simply installing a camera aimed at the food in the production line, quality control can be quickly integrated into existing production environments without causing delays in the production process. This setup makes it possible to detect defects in real-time, which improves the quality of food production without affecting the throughput speed 20. In one implementation form, the data displayed on the graphical user interface comprises one or a combination of the following data types: a timeline of identified defective food products, an overview of the number of identified defects 25 subdivided based on defect characteristics,and a live view of the camera image with indicated defects. This variation in data types offers multiple benefits. The timeline of identified defects makes it possible to analyze defect patterns, allowing operators to gain insight into recurring problems and respond quickly to quality issues. The overview of the 30 defects subdivided by properties provides detailed information about common defect types, which contributes to targeted improvements in the production process. The live view of the camera image with indicated defects offers immediate visual feedback and enables operators to take immediate action,which further increases the speed and effectiveness of quality control. Together, these data types offer extensive and flexible support for quality management within the production line. 2024 / 5836 BE2024 / 5836 15 In one implementation form, the device comprises at least two cameras that capture substantially the same food products in the production line from different perspectives. This offers the advantage that food products are observed from multiple angles, allowing the AI ​​model to identify defects that might not be fully visible from a single perspective. By combining multiple views, hidden or partially covered defects are better detected, which significantly improves the accuracy of quality control. Additionally, this approach increases the reliability of defect detection under variable product positions or orientations in the production line, because the defects remain in view even when products rotate or move. This multi-camera configuration therefore increases the robustness and accuracy of the device.without the production line losing speed. In what follows, the invention is described by means of non-limiting examples that illustrate the invention, and which are not intended or should not be interpreted to limit the scope of the invention. EXAMPLES A digital 3D basic model of a food product can be made based on 20 photos taken from different angles and perspectives of the product. The product is photographed from multiple positions around the object to capture all sides and details, whereby the photos offer small variations in angle and position. For food products with a more complex shape, smaller variations in angle are taken and more photos per angle. Subsequently, these photos are fed into 25 specialized photogrammetry software, which uses algorithms to identify and triangulate common points in the photos,resulting in a three-dimensional point cloud of the surface. Based on this point cloud, a mesh is created that represents the shape of the product in 3D. 30 After generating the digital 3D base model of the food product, the dimensions and shapes of this base model are varied to obtain several digital models that represent realistic variations of the original product. These variations can be performed manually or automated using algorithms, depending on the required quantity and complexity of the models. 2024 / 5836 BE2024 / 5836 16 In a manual execution, an operator adjusts the dimensions and shapes of the base model within 3D modeling software, allowing for full control over variations. For larger datasets, algorithms are actually more effective because they can apply changes quickly and consistently, which leads to a more efficient production of realistic model variants.5 Parametric modeling can also be used to determine basic parameters such as height,width and depth can be flexibly adjusted within predefined margins. This automatically generates variants of the base model. For example, the algorithm can increase or decrease the height of the model within a 10% margin while maintaining proportions, creating realistic variations. PerlinNoise or SimplexNoise algorithms can also be used to introduce subtle asymmetrical distortions that mimic natural irregularities on the surface. These algorithms add noise to specific points on the model, such as slight bulges or indentations, resulting in realistic-looking variations that often occur in food products. A Monte Carlo algorithm can be applied to select arbitrary variations within 20 predefined boundary values. At each iteration, the algorithm selects new values ​​for the dimensions, creating a dataset with diverse models without repetition of the same variations. This is effective for quickly generating a wide range of shapes. 25 Finally, Gaussian scaling can be used,where subtle fluctuations around an average value are applied. This algorithm ensures that most variations lie close to the base model, while a few larger deviations are added for diversity. As a result, Gaussian Scaling simulates subtle natural variations in size and shape.30 To add realistic surface characteristics to the digital models, various properties such as texture, color, reflectivity, and roughness can be adjusted to make the model visually correspond better to a real food product. Various algorithms and techniques are employed to simulate these variations accurately and realistically.35 First, textures and roughness can be created using bump mapping and normal mapping, techniques that add details to the surface without changing the underlying geometry. With bump mapping, for example, a slightly rough texture can be created. be added, simulating a granular surface,while normal mapping provides more depth in the texture by adjusting the fall of light. This is particularly useful for simulating fine structures such as small ridges or irregularities on the product surface.5 For applying color variations, PerlinNoise or SimplexNoise algorithms can be applied, which generate subtle differences in color and simulate natural color gradations, such as light discolorations or irregular shades. These algorithms add noise to the color pattern, creating a realistic, non-uniform color distribution. In addition, shader algorithms can be used to flexibly adjust the intensity and saturation of colors, so that the model resembles the natural color variations of the food product. To mimic reflectivity and gloss, specular mapping can be applied. This algorithm makes it possible to adjust the reflectivity of specific parts of the surface,so that parts of the product acquire a slight sheen while other noble parts retain a matte finish. This is useful for products that have different material structures on a single surface, such as glossy and off-white spots, which contributes to a more realistic surface. For products that are partially transparent or where depth effects are desired, subsurface scattering and opacity mapping can be applied. Subsurface scattering simulates how light spreads beneath the surface of a product, which gives a softer and more natural effect on semi-transparent surfaces, such as on thin25 peel or juicy parts of the product. Opacity mapping is used to simulate partial translucency, for example on thinner edges or parts of the product that are less dense. By applying a combination of these techniques, a versatile and realistic dataset can be created, which helps the AI ​​model to accurately identify defects under various conditions. Each algorithm contributes to strengthening the visual accuracy of the digital models,whereby the model is effectively trained on defect recognition that reflects realistic variations in surface characteristics.35 To generate 3D defects that represent realistic variants of actual damage to food products, use can be made of 2024 / 5836 BE2024 / 5836 18 different techniques and sources. The process begins with collecting imagery of the defects that actually occur in food products on the production line. This imagery serves as a reference to capture the various characteristics of the defects, such as shape, size, texture, color changes, and intensity of the damage. Defects such as cracks, discolorations, bruises, and superficial damage can be analyzed in as much detail as possible. Based on this data, digital models of the defects are generated that represent each of the captured characteristics. In this way, cracks can be modeled with specific lengths, depths, and texture details, so that the different 10 types of cracks that can occur in food products,be realistically simulated. For color variations, a model can be built in which color gradations and distribution patterns are realistically displayed, based on color information and distribution patterns observed in the reference image material.15 An example of automatically generating a 3D model of a defect based on a photo begins with the analysis of an image in which a specific defect, such as a crack, is visible. Software sample recognition can be used to identify the edges, length, and depth indicators of the crack in the photo.20 This information is then converted into depth and texture data, with which a digital 3D model of the crack can be built. Algorithms such as edge detection mark the contours of the crack, while depth mapping captures the depth profiles. These profiles are converted into a three-dimensional mesh that accurately represents the shape and depth of the crack. To create a realistic texture25 to add, can the surface of the 3D model be edited with a bump map or displacement map,in which noise is added to simulate the irregularities and rough edges of the crack as observed in the original photo. This method makes it possible to generate 3D models of various defects in an automated manner, based on actual damage patterns that are captured in imagery. The use of algorithms such as PerlinNoise and Gaussian distribution is useful for generating variations within each type of defect. These algorithms add subtle random variations to the properties of the defects, such as the size, the intensity of the color change, and the texture of the damage. This ensures that the generated defects are not uniform, but show natural variation that corresponds to what can occur in reality. For example, PerlinNoise can add small, irregular variations to the edges of a crack,while a Gaussian distribution is used to realistically decrease the intensity of a discoloration towards the edges. In this way, a varied dataset of the defects is created with natural variations that are representative of what can occur in reality.5 Additionally, a dataset of the defects can be expanded by using rigid body simulation to counteract standard deformations such as dents. In this case, a rigid body simulation is used to simulate the exact shape and depth of a dent on a standalone model,where the effects of forces such as pressure or impact on the surface are simulated. This resulting model of the dent can then be saved as a standard so that the defects can be applied to various food product models without the need to perform the full simulation again each time. The dataset of digital models of food products can be combined with the dataset of digital models of the defects to generate a realistic and varied set of digital models of food products with the defects. This process begins with selecting a digital model of a food product and choosing a suitable defect model from the defect dataset, such as a crack, dent,discoloration or superficial damage. Each defect model can be specifically positioned and scaled to accurately mimic the natural occurrence of defects in the food product. The defects can be placed at specific locations on the food model to create variation. This placement can be set manually or automated by an algorithm that applies the defects at random or specific positions. By applying random positions and introducing variations in the orientation or size of the defect model, a dataset is created with naturally diverse configurations of the food product with the defects. For example, defects such as rotten spots, mold spots, or other forms of internal degradation are placed randomly on the food product because they can occur anywhere in practice, and defects such as impact damage are often found on the ends, corners, or edges of a food product,because these areas are more vulnerable to physical impact.35 An example of this process is combining a basic model of a food product with a crack model, where the crack can be stretched, rotated, or scaled to match the surface and shape of the food product. The digital model of the defect can be scaled to the size of the digital model of the food product. Algorithms such as random positioning or Gaussian distribution can be applied to adjust the size, placement, and orientation of the defect in a controlled, variable manner,5 so that the defect looks more realistic on different versions of the food product model. After combining the food product and the defect, texture adjustment can be performed to soften the transition between the food surface and the defect.10 By using bump mapping, normal mapping or displacement mapping, the defect can be seamlessly integrated into the surface of the food product,whereby it looks less like an “added” element and more like a natural defect that originates in the product itself. 15 To simulate a production environment around the digital models of food products with the defects, a 2D or 3D simulation can be set up, depending on the desired degree of realism and detail. In both cases, the goal of the simulation is to replicate a realistic inspection environment, based on imagery and observations of the actual production environment. 20 In a 3D simulation, a virtual environment is created around the three-dimensional models of food products with the defects. The process begins with analyzing imagery of the actual production environment, such as videos or photos of the production line. This imagery is used to analyze relevant details such as the arrangement of conveyor belts, equipment,to capture lighting and environmental objects. Based on this data, a 3D model of the production environment can be built that forms a more accurate representation of the physical environment. In this process, the relative positions of equipment, conveyor belts, supporting structures, and other elements can be simulated, so that the food products are situated in a realistic context. The simulation can mimic various aspects of the production environment to train the AI ​​model in recognition under varying conditions. For example, variable lighting can be added, such as strong lamps or natural light falling through windows, with which realistic shadows and light reflections are simulated. The light incidence and intensity can be set variably to simulate different times of the day,such as the influence of morning or evening light on the conveyor belt 2024 / 5836 BE2024 / 5836 21 or the bright artificial light that illuminates the textures of the food product. This helps the AI ​​model to recognize defects under various lighting conditions. Variable obstacles and personnel can also be added to the simulation. Realistic shadows and reflections can further be added based on lighting and the presence of equipment and personnel in the environment. By simulating, for example, the reflection of metal parts of a machine or the shadow pattern of a moving arm, the model is prepared for the visual inferences that are typical in a production environment. 10 In a 2D simulation, a comparable production environment is simulated around two-dimensional images of the digital models of food products with defects. Here, the images of the food product are placed on a 2D background that visually represents the environment,such as an image of a conveyor belt or sorting system. Lighting and shadows can also be applied in this 2D simulation15 by means of filters or shaders that generate different light intensities and shadow effects. It is possible to place multiple digital models of food products, both metals without defects, in a single simulated production environment to create a20 realistic image of the production line. This setup simulates, for example, the configuration of food products on a conveyor belt, where products are displayed in bulk as would occur in a real production line. By placing the digital models at specific positions, a detailed representation is created in which products partially overlap or lie at different angles and positions25. This simulation of a static image of the production environment ensures that the AI ​​model can be trained on various product configurations, including variations in the visibility of defects,shadowing and lighting effects that further enhance realism. 30 In a subsequent step, physical interactions between the digital models of food products can be simulated to imitate the dynamics of a production line. Algorithms such as rigid body dynamics and collision detection are used here to mimic the effect of contact between products. These algorithms calculate, for example, how products slide, rotate, or overlap on the conveyor belt when they come into contact with each other or with other objects in the environment. Upon collision, slight deformations, tilts, or shifts 2024 / 5836 BE2024 / 5836 22 may occur, depending on the position and direction of the contact, which provides a realistic picture of how products behave towards each other on the production line. With this setup, an animated simulation can be generated, in which the movement of food products through the simulated production line is captured as a video output5. This dynamic video simulates the movement on the conveyor belt, while the environmental elements,such as light and shadow, adapting to the movements. The next step is obtaining two-dimensional images of the digital models of food products in the production environment. Here, it is important to generate a wide variety of images that accurately simulate the possible perspectives of the camera in the actual production environment. In a 3D simulation, virtual cameras are used to capture the models from 15 different angles and positions. The camera can be positioned in a way that corresponds to the actual cameras in the production line, so that the angles of incidence and perspectives in the images correspond to the realistic inspection angle. By using camera rotation and zoom functions, a wide range of perspectives can be captured, from top views and 20 side views to slight oblique angles. In addition, variable lighting settings can be used to adjust the lighting in the 3D environment,so that the defects are visible under various lighting conditions. This method yields a wide variety of images that simulate a lifelike inspection environment. In a 2D simulation, it is possible to zoom in on specific areas and apply filters and perspective distortions to display the images from different angles and positions, mimicking the angles and positions of a typical inspection camera on the production line. Lighting effects can also be added to the 2D simulation, for example by applying exposure filters to vary the intensity and direction of the light in the images. This creates a set of two-dimensional images that simulate various inspection angles and lighting conditions, comparable to the deco,