Vision-Based System for Food Ingredient Weight Estimation and Inedible-Part Sorting

Abstract

The present invention relates to a vision-based food raw material weight prediction and non-edible part sorting system. It utilizes images of food raw materials to precisely segment and detect non-edible and diseased parts, calculates the weight of edible parts in real time based on the segmented results, quantitatively determines the quality of the food raw materials using the calculated information, and enables the automatic removal of non-edible parts when necessary. The present invention comprises: an image acquisition unit for acquiring an image of a food raw material; a deep learning-based segmentation network for segmenting and detecting inedible parts and diseased parts of the food raw material from the acquired image; a weight prediction unit for calculating the weight of an edible part of the food raw material using the segmentation detection results; and a control unit for generating quality judgment information of the food raw material based on the output of the segmentation network and the weight prediction unit.

Description

Technology Field

[0001] The present invention relates to a vision-based food raw material weight prediction and non-edible part sorting system, and more specifically, to a vision-based food raw material weight prediction and non-edible part sorting system configured to automatically segment and detect non-edible parts and diseased parts using images of food raw materials and to quantitatively predict the weight of edible parts. Background Technology

[0002] Generally, the quality of food raw materials varies significantly during pretreatment processes such as harvesting, washing, and processing, depending on factors such as external damage, the extent of disease occurrence, and the presence of inedible parts; these quality factors directly affect the safety and marketability of the final product.

[0003] In particular, irregular food ingredients such as chili peppers, green onions, mushrooms, and onions present a problem in that accurate quality assessment is difficult using human visual inspection or simple standard-based inspection methods due to their irregular size, shape, curvature, and color. Therefore, there is a growing need for technology to consistently evaluate the condition of food ingredients and quantify the amount of edible parts.

[0004] Conventional quality determination technology for food ingredients relies primarily on visual inspection by operators or simple color and area-based classification methods, which makes it difficult to identify diseased areas with fine details, limits the ability to clearly distinguish between inedible and edible parts, and fails to provide a function to evaluate edible weight in real time.

[0005] Furthermore, in conventional technology, disease detection, weight prediction, and inedible part removal functions are often implemented as separate individual systems, making it difficult to automate the entire process and limiting its application in mass production environments due to insufficient real-time processing speed.

[0006] Furthermore, existing deep learning-based classification models can be used for determining grades at the whole raw material level, but they have limitations in that they are difficult to perform quantitative analysis functions such as precise segmentation at the pixel level or edible weight prediction. The problem to be solved

[0007] Accordingly, the present invention is proposed to resolve the aforementioned conventional problems. The objective of the present invention is to provide a vision-based food raw material weight prediction and inedible part sorting system that uses images of food raw materials to precisely segment and detect inedible parts and diseased parts, calculates the weight of edible parts in real time based on the segmented results, quantitatively determines the quality of the food raw materials using the calculated information, and automatically removes inedible parts when necessary. means of solving the problem

[0008] To achieve the above objectives, the vision-based food raw material weight prediction and non-edible part sorting system according to the technical concept of the present invention is characterized by the following technical configurations: the vision-based food raw material weight prediction and non-edible part sorting system for determining the quality of food raw materials includes an image acquisition unit for acquiring an image of a food raw material, a deep learning-based segmentation network for segmenting and detecting non-edible parts and diseased parts of the food raw material from the acquired image, a weight prediction unit for calculating the weight of edible parts of the food raw material using the segmentation detection results, and a control unit for generating quality judgment information based on the outputs of the segmentation network and the weight prediction unit.

[0009] Here, the deep learning-based segmentation network may be characterized by including a segmentation model for segmenting inedible parts or diseased parts of food ingredients into pixel units.

[0010] In addition, the segmentation model may be characterized by including a YOLO-based structure and further including an MCEB or attention-based module for extracting multi-scale features.

[0011] In addition, the weight prediction unit may be characterized by including a regression model that calculates the edible weight using one or more of the pixel area corresponding to the divided edible part, the circumscribed rectangle size, shape-based features, and color features as input values.

[0012] In addition, the regression model may be characterized by including one or more of a multiple linear regression model or a lightweight deep learning regression neural network.

[0013] In addition, the control unit may be characterized by calculating the location of the inedible part from the result of the segmented network and generating a control signal to remove the inedible part based on the location information.

[0014] In addition, the removal of the above-mentioned inedible part may be characterized by being performed using one or more of the following methods: a cutting method using a cutting blade mounted on a multi-joint robot arm, a removal method using an air injection nozzle, or a removal method using a suction device.

[0015] In addition, the deep learning-based partitioning network described above may be characterized by being converted to ONNX-based and configured to enable real-time inference through the TensorRT inference engine.

[0016] In addition, the image acquisition unit may be characterized by including an RGB camera for photographing food raw materials moving along a conveyor and an LED lighting device for illuminance uniformity.

[0017] In addition, the control unit may be characterized by performing one or more functions among classifying food raw materials into grades, sorting, storing data, or providing notifications based on quality judgment information.

[0018] Meanwhile, the vision-based food raw material weight prediction and non-edible part selection method of the present invention is characterized by the technical configuration of the steps of acquiring an image of a food raw material, segmenting and detecting non-edible parts and diseased parts of the food raw material using a deep learning-based segmentation network, calculating the weight of edible parts from the segmentation detection results, and generating quality judgment information.

[0019] Here, the segmentation detection step may be characterized by being performed using a segmentation model including a YOLO-based structure and an MCEB or attention-based module.

[0020] In addition, it may be characterized by further including a step of performing one or more of grading, sorting, or removing inedible parts of food raw materials based on the above quality judgment information. Effects of the invention

[0021] The vision-based food raw material weight prediction and non-edible part sorting system according to the present invention has the effect of precisely segmenting and detecting non-edible parts and diseased parts using a deep learning-based segmentation network on an acquired image of a food raw material.

[0022] In addition, the vision-based food raw material weight prediction and non-edible part sorting system according to the present invention has the effect of being able to calculate the edible weight of food raw materials in real time by applying a regression model that takes area, shape, and color information of the divided edible parts as input.

[0023] Furthermore, the vision-based food raw material weight prediction and inedible part sorting system according to the present invention can quantitatively determine the quality of food raw materials based on the division results and weight calculation results, and has the effect of automatically removing inedible parts by selectively driving a cutting module, an air injection device, or a suction device according to the operation of the control unit. Moreover, by applying a high-speed inference structure based on ONNX transformation and TensorRT, the vision-based food raw material weight prediction and inedible part sorting system according to the present invention enables real-time processing even on production lines where large quantities of food raw materials are continuously supplied, thereby having the effect of significantly improving the automation and efficiency of the entire process. Brief explanation of the drawing

[0024] FIG. 1 is a block diagram illustrating the overall configuration of a vision-based food raw material weight prediction and non-edible part sorting system according to an embodiment of the present invention. FIG. 2 is a reference diagram for explaining the configuration of an image acquisition unit for acquiring an image of a food raw material in an embodiment of the present invention, and illustrates the arrangement of an RGB camera and a lighting structure. FIG. 3 is a configuration diagram for explaining the internal structure of a deep learning-based partitioning network (MCEB-YOLO) applied to an embodiment of the present invention. FIG. 4 is a reference diagram for explaining the division results of inedible parts and diseased parts of food raw materials according to an embodiment of the present invention, and illustrates the comparison results of various model variations (YOLOv8, TlCoslr, AugAM, etc.). Figure 5 is a reference diagram for explaining the segmentation performance corresponding to diseased parts (spouse, mold, reddening, etc.) of food raw materials according to an embodiment of the present invention, and shows the comparison results between the YOLO-based model and the MCEB-YOLO model. FIG. 6 is a flowchart illustrating a procedure for predicting the weight of a food ingredient according to an embodiment of the present invention, showing the steps of food ingredient recognition, segmentation, feature extraction, application of a regression model, and weight calculation. FIG. 7 is a reference diagram for explaining the morphological characteristics of food raw materials used in embodiments of the present invention and the definition of inedible parts (calliforms, stems, etc.), and shows the original PCD, depth-based PCD, and Pericap / Calyx annotation results. FIG. 8 is a block diagram for explaining the configuration of a non-edible part removal unit according to an embodiment of the present invention, illustrating the operational relationship between a control unit, a cutting module, an air injection device, and a suction device. FIG. 9 is a reference diagram for explaining the overall flow of a vision-based food raw material weight prediction and non-edible part selection method according to an embodiment of the present invention, illustrating the sequence of image acquisition, segmentation, weight prediction, quality determination, and removal / selection steps. FIG. 10 is a reference diagram for explaining the performance of a 3D point cloud-based segmentation model applied in an embodiment of the present invention, and shows the results of comparison with the Original PCD, Ground truth, and various models (PointNet, PointNet++, SE-PointNet++, AE-PointNet++). Specific details for implementing the invention

[0025] A vision-based food raw material weight prediction and non-edible part sorting system according to embodiments of the present invention will be described in detail with reference to the attached drawings. As the present invention is susceptible to various modifications and may take various forms, specific embodiments are illustrated in the drawings and described in detail in the text. However, this is not intended to limit the present invention to the specific disclosed forms, and it should be understood that the invention includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention. Similar reference numerals have been used for similar components in the description of each drawing. In the attached drawings, the dimensions of structures are depicted enlarged or reduced to the actual size to ensure clarity of the invention or to allow for a schematic understanding of the configuration.

[0026] Additionally, terms such as "first," "second," etc., may be used to describe various components, but said components should not be limited by said terms. These terms are used solely for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be named the second component, and similarly, the second component may be named the first component. Meanwhile, unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application.

[0027] <Example>

[0028] FIG. 1 is a block diagram illustrating the overall configuration of a vision-based food raw material weight prediction and non-edible part sorting system according to an embodiment of the present invention.

[0029] As described above, the vision-based food raw material weight prediction and non-edible part selection system according to an embodiment of the present invention comprises an image acquisition unit (110), a deep learning-based segmentation network (120), a weight prediction unit (130), a control unit (140), and a non-edible part removal unit (150) as main components, and is configured to automatically segment non-edible parts and diseased parts from an image of a food raw material, quantitatively predict the weight of edible parts, determine the quality of the food raw material based on the prediction results, and automatically remove non-edible parts when necessary.

[0030] A system according to an embodiment of the present invention captures food raw materials moving on a high-speed conveyor in real time to obtain high-resolution images, and processes the obtained images through a deep learning-based segmentation network to precisely segment and detect inedible parts such as mold, reddening, calyx, and stem. For the segmented edible areas, a weight prediction unit extracts features such as area, shape, and color and applies a regression-based weight prediction model to calculate the actual edible weight. A control unit determines the quality of the food raw materials by synthesizing the segmentation and prediction results, and generates a control signal to automatically remove inedible parts by cutting, air injection, or suction if necessary.

[0031] Hereinafter, a vision-based food raw material weight prediction and non-edible part sorting system according to an embodiment of the present invention will be described in detail, focusing on the above-mentioned major components.

[0032] An image-based food raw material weight prediction and non-edible part selection system according to an embodiment of the present invention is configured such that a plurality of modules are interconnected to quantitatively determine the quality of food raw materials, and includes an image acquisition unit (110), a deep learning-based segmentation network (120), a weight prediction unit (130), a control unit (140), and a non-edible part removal unit (150) as shown in FIG. 1. Each of the above components is configured to perform a series of operations to segment and detect non-edible parts and diseased parts from an image of a food raw material, calculate the weight of edible parts, determine quality through the calculated information, and remove non-edible parts if necessary.

[0033] The image acquisition unit (110) above serves to acquire images of food raw materials. The image acquisition unit (110) includes an RGB camera for acquiring images of food raw materials moving along a conveyor in real time and an LED lighting device for uniform illumination, and is configured to enable stable image acquisition in various shooting environments as shown in FIG. 2. By providing multiple shooting conditions such as Front View, Close-up View, and Side View, the accuracy of deep learning-based analysis can be increased, and the image acquisition unit (110) is a core component that provides base data for the segmentation detection and weight calculation processes performed in subsequent steps.

[0034] The deep learning-based segmentation network (120) described above serves to segment and detect inedible parts and diseased parts of food raw materials from an acquired image. The deep learning-based segmentation network (120) includes a YOLO-based structure and is configured to further include a Multi-scale Convolutional Enhancement Block (MCEB) or an attention-based module for extracting multi-scale features. As shown in FIG. 3, the deep learning-based segmentation network (120) extracts features of various resolution levels from an input image and strengthens the correlation between the extracted features, thereby enabling precise segmentation of inedible parts such as fine diseased parts on the surface of food raw materials, calyxes, and stems at the pixel level. As shown in FIG. 4 and FIG. 5, the deep learning-based segmentation network (120) improves the accuracy of segmentation boundaries and the ability to detect diseased areas compared to existing YOLO models, and can secure high precision even when compared to various modified models. This composition has the advantage of enabling the stable separation of inedible parts even in environments where there are diverse changes in the appearance of food ingredients or changes in surface reflection and color.

[0035] The weight prediction unit (130) serves to calculate the weight of the divided edible parts. The weight prediction unit (130) includes a regression model that calculates the edible weight by taking one or more of the pixel area corresponding to the edible part, circumscribed rectangle size, shape-based features, and color features as input values ​​from the output of the deep learning-based segmentation network (120). As illustrated in FIG. 6, the weight prediction unit (130) performs a feature extraction step based on the edible region segmented in the image of the food raw material, and inputs the extracted features into the regression model to calculate a predicted value corresponding to the actual weight. The regression model can be implemented as one or more of a multiple linear regression (MLR) model or a lightweight deep learning regression neural network, which provides the advantage of maintaining high prediction accuracy while satisfying real-time processing requirements. In addition, the original PCD, depth-based PCD, and pericap / calyx annotation examples shown in FIG. 7 demonstrate that quantitative analysis based on the morphological characteristics of food ingredients is possible, which supports the fact that the weight prediction unit (130) can perform reliable predictions even in various forms of food ingredients.

[0036] The control unit (140) plays the role of generating quality judgment information for food raw materials based on the output of the deep learning-based segmentation network (120) and the weight prediction unit (130). The control unit (140) calculates the location of the inedible part from the segmentation detection result and generates a control signal to control the inedible part removal unit (150) based on the calculated location information. In addition, the control unit (140) can perform functions such as classifying the grade of the food raw material, sorting, storing data, or providing notifications by referring to the edible weight calculated by the weight prediction unit (130), which plays an important role in quality control of food raw materials and automated sorting processes. As shown in FIG. 8, the control unit (140) selectively operates one or more of the cutting module (151), the air injection device (152), and the suction device (153) to perform a series of control operations to remove the inedible part.

[0037] The above-mentioned inedible part removal unit (150) serves to remove inedible parts of food raw materials. The inedible part removal unit (150) includes one or more of a cutting module (151), an air injection device (152), or a suction device (153), and removes the corresponding part by operating the corresponding device according to the location information of the inedible part provided by the control unit (140). The cutting module (151) can remove inedible parts by a direct cutting method including a blade or cutting edge, and the air injection device (152) can remove small diseased parts or weak inedible parts by a non-contact method using a high-pressure air jet. The suction device (153) is useful for removing inedible parts with low adhesion, such as mold or surface contaminants, by using suction force. This configuration allows for the selection of an optimal removal method depending on process conditions and the type of inedible part, and provides the effect of simultaneously ensuring high processing speed and hygiene in an automated food raw material sorting process.

[0038] FIG. 9 illustrates the overall flow of an image-based method for predicting the weight of food raw materials and sorting inedible parts, wherein the image acquisition step (S210), the inedible part and disease segmentation step (S220), the weight prediction step (S230), the quality determination step (S240), and the removal or sorting step (S250) proceed sequentially. This flow corresponds to the method components defined in the claims, and the series of procedures performed at the system level are clearly indicated, making it easy to understand the overall operation process.

[0039] Figure 10 illustrates a comparison of analysis results based on 3D point clouds and shows the performance differences between the Original PCD, Ground truth, and various segmentation models (PointNet, PointNet++, SE-PointNet++, AE-PointNet++, Ours). This suggests that in the embodiment of the present invention, weight prediction and feature extraction can be extended not only to 2D images but also to 3D point cloud data, and demonstrates that stable data analysis is possible even when there is significant morphological variation in food raw materials.

[0040] As described above, the embodiment of the present invention provides a configuration that is highly advantageous for quantitatively evaluating the quality of food raw materials and automating the production process by integrating functions for separating inedible parts, weight prediction, and automatic removal throughout the entire image processing process of food raw materials.

[0041] A vision-based method for predicting the weight of food raw materials and sorting out inedible parts according to an embodiment of the present invention comprises, as illustrated in FIG. 9, a step of acquiring an image of a food raw material (S210); a step of segmenting and detecting inedible parts and diseased parts of the food raw material from the acquired image using a deep learning-based segmentation network (S220); a step of calculating the weight of the edible parts of the food raw material from the segmentation detection results (S230); and a step of generating quality judgment information of the food raw material based on the segmentation detection results and the weight calculation results (S240). The method further comprises a step of removing or sorting out inedible parts as necessary (S250). Each of the above steps will be explained in more detail below with reference to FIG. 9.

[0042] The step of acquiring an image of the food raw material (S210) serves to photograph the food raw material moving on the conveyor in real time using an image acquisition unit (110). As shown in FIG. 2, the image acquisition unit (110) is configured to secure an image of uniform quality even under various viewpoints and lighting conditions by including an RGB camera and a lighting device, and the acquired image is used as input for segment detection and weight prediction in subsequent steps. This image acquisition step is a process of providing basic data that determines the reliability and accuracy of the entire method by enabling the acquisition of data that sufficiently reflects the morphological changes, color distribution, and surface defects of the food raw material.

[0043] The step (S220) of segmenting and detecting inedible parts and diseased parts of the above food raw material using a deep learning-based segmentation network applies a deep learning-based segmentation network (120) to the acquired image to segment diseased parts, such as mold, reddening, and decayed areas, and inedible parts, such as stems or calyxes, present on the surface of the food raw material in pixel units. The deep learning-based segmentation network (120) is implemented in a form in which an MCEB or attention-based module is combined with a YOLO-based structure as illustrated in FIG. 3, and this structure provides high-precision segmentation performance corresponding to various sizes and shapes of food raw materials. As exemplified in FIG. 4 and FIG. 5, it can be confirmed that the segmentation network of the present invention has superior boundary recognition ability and diseased area detection ability compared to various model variations, and the segmentation detection step performs a core function that serves as the basis for subsequent weight prediction and quality judgment.

[0044] The step (S230) of calculating the weight of the edible part of the food ingredient described above involves the weight prediction unit (130) performing feature extraction and regression prediction based on the area corresponding to the edible part among the output results of the deep learning-based segmentation network (120). The weight prediction unit (130) takes the segmented edible area as input as illustrated in FIG. 6, extracts pixel area, circumscribed rectangle size, shape-based features, color features, etc., and applies the extracted features to a regression model to quantitatively calculate the weight of the edible part. The regression model may be composed of a multiple linear regression model or a lightweight deep learning regression neural network, which enables the generation of a predicted value close to the actual weight while minimizing computational load. Furthermore, the chili pepper shape classification image and pericap / calyx annotation information illustrated in FIG. 7 demonstrate that the weight prediction is a highly reliable analysis that reflects morphological and structural features, rather than being limited to simple area-based calculations.

[0045] The step of generating the quality judgment information (S240) involves the control unit (140) determining the quality of the food raw material by synthesizing the results of the segmentation detection step (S220) and the weight prediction step (S230). The control unit (140) can classify the grade of the food raw material or select raw materials that do not meet quality standards by analyzing the location and area of ​​the segmented inedible parts, and, if necessary, can perform yield analysis or production history data storage functions based on the edible weight. The quality judgment information generated in this step is utilized as an operation control signal for the inedible part removal unit (150), which provides the effect of significantly improving the efficiency of the automated sorting process.

[0046] The step (S250) of removing or sorting the inedible parts described above is performed by driving one or more of the cutting module (151), air injection device (152), or suction device (153) included in the inedible part removal unit (150) according to the command of the control unit (140) to remove the inedible parts or sort the food raw materials. FIG. 8 illustrates the configuration and mutual operation relationship of each removal method, and the cutting method is suitable for separating inedible parts of a certain size or larger, the air injection method is useful for removing small surface diseased parts, and the suction method is effective for removing surface contaminants. These steps contribute to reducing quality deviations by increasing the level of automation in the entire process and minimizing worker intervention.

[0047] The vision-based food raw material weight prediction and non-edible part sorting method described above can rapidly and accurately evaluate the quality of food raw materials by consistently performing all procedures from segmentation, prediction, judgment, and removal based on the acquired images of food raw materials. Furthermore, when applied in conjunction with the 3D-based analysis process shown in Fig. 10, high reliability can be secured even for food raw materials with significant morphological diversity. In addition, this method can operate stably even in a real-time processing environment, so it can be effectively applied to automated food processing lines.

[0048] Although preferred embodiments of the present invention have been described above, the present invention may use various variations, modifications, and equivalents. It is clear that the present invention can be applied in the same way by appropriately modifying the above embodiments. Therefore, the above description does not limit the scope of the present invention, which is defined by the limitations of the following claims. Explanation of the symbols

[0049] 110: Image acquisition unit 120: Deep learning-based segmentation network 130: Weight prediction unit 140: Control unit 150: Non-edible part removal section 151: Cutting module 152: Air injection device 153: Intake device

Claims

Claim 1 A vision-based food raw material weight prediction and non-edible part sorting system for determining the quality of food raw materials, comprising: an image acquisition unit for acquiring an image of a food raw material; a deep learning-based segmentation network for segmenting and detecting non-edible parts and diseased parts of the food raw material from the acquired image; a weight prediction unit for calculating the weight of edible parts of the food raw material using the segmentation detection results; and a control unit for generating quality determination information of the food raw material based on the outputs of the segmentation network and the weight prediction unit. Claim 2 A vision-based food raw material weight prediction and inedible part selection system according to claim 1, characterized in that the deep learning-based segmentation network includes a segmentation model for segmenting inedible parts or diseased parts of food raw materials in pixel units. Claim 3 A vision-based food raw material weight prediction and non-edible part selection system according to claim 2, characterized in that the segmentation model includes a YOLO-based structure and further includes a Multi-scale Convolutional Enhancement Block (MCEB) or an attention-based module for extracting multi-scale features. Claim 4 A vision-based food raw material weight prediction and non-edible part sorting system according to claim 1, wherein the weight prediction unit comprises a regression model that calculates edible weight using one or more of a pixel area corresponding to a divided edible part, circumscribed rectangle size, shape-based features, and color features as input values. Claim 5 A vision-based food raw material weight prediction and non-edible part sorting system according to claim 4, characterized in that the regression model includes one or more of a multiple linear regression (MLR) model or a lightweight deep learning regression neural network. Claim 6 A vision-based food raw material weight prediction and inedible part sorting system according to claim 1, wherein the control unit calculates the location of the inedible part from the result of the segmentation network and generates a control signal for removing the inedible part based on the location information. Claim 7 A vision-based food raw material weight prediction and non-edible part sorting system according to claim 6, characterized in that the removal of the non-edible part is performed using one or more of the following methods: a cutting method using a cutting blade mounted on a multi-joint robot arm, a removal method using an air injection nozzle, or a removal method using a suction device. Claim 8 A vision-based food raw material weight prediction and non-edible part sorting system according to claim 1, characterized in that the deep learning-based segmentation network is converted to ONNX-based and configured to enable real-time inference through a TensorRT inference engine. Claim 9 A vision-based food raw material weight prediction and non-edible part sorting system according to claim 1, wherein the image acquisition unit comprises an RGB camera for photographing food raw materials moving along a conveyor and an LED lighting device for illuminance uniformization. Claim 10 A vision-based food raw material weight prediction and non-edible part sorting system according to claim 1, wherein the control unit performs one or more functions among grade classification, sorting, data storage, or notification provision of food raw materials based on quality judgment information. Claim 11 A vision-based method for predicting the weight of a food ingredient and sorting inedible parts for determining the quality of a food ingredient, comprising: a step of acquiring an image of a food ingredient; a step of segmenting and detecting inedible parts and diseased parts of the food ingredient from the acquired image using a deep learning-based segmentation network; a step of calculating the weight of the edible parts of the food ingredient from the segmentation detection results; and a step of generating quality determination information of the food ingredient based on the segmentation detection results and the weight calculation results. Claim 12 A vision-based food raw material weight prediction and non-edible part selection method according to claim 11, characterized in that the segmentation detection step is performed using a segmentation model including a YOLO-based structure and a MCEB (Multi-scale Convolutional Enhancement Block) or an attention-based module. Claim 13 A vision-based food raw material weight prediction and inedible part sorting method according to claim 11, further comprising the step of performing one or more of grade classification, sorting, or removal of inedible parts of food raw materials based on the quality judgment information.

Images