A multi-stage detection system and method for beef product processing

Abstract

The application discloses a multi-stage detection system and method for beef product processing and relates to the technical field of multi-source data processing; the method comprises the following steps: acquiring target gas of a target beef product; sucking the target gas into a common chamber of a double sensor; performing first detection on the gas in the common chamber to obtain an environmental dry coefficient value; if the environmental dry coefficient value is greater than a preset dry coefficient value, performing second detection on the target gas to obtain a gas acidity spectrum, substituting the gas acidity spectrum into a neural network model to perform identification to obtain an acidity sample probability; if the acidity sample probability is less than or equal to a preset acidity threshold value, acquiring image data of the target beef product, performing third detection on the image data to obtain a meat color result; and if the acidity sample probability is greater than the preset acidity threshold value, issuing an acid spoilage signal by a terminal. The application detects through smell and image, improves the precision of beef quality detection, and reduces false detection.

Description

Technical Field

[0001] This invention belongs to the field of multi-source data processing technology, specifically relating to a multi-level detection system and method for beef product processing. Background Technology

[0002] In the beef product processing sector, this technology integrates and analyzes multi-dimensional data collected from the entire production process to ensure effective product quality testing and safety control. Multi-source data processing technology utilizes a series of data acquisition and analysis facilities, such as sensor terminals, image recognition modules, IoT gateways, and data processing servers, to transform scattered production, testing, and traceability data into standardized information usable for multi-level testing. This forms a multi-dimensional testing and analysis network that provides decision support to the production process. This provides efficient and precise technical assurance for the safe production and quality control of beef products.

[0003] Existing technology (publication number: CN118604200A) discloses a method for detecting azo synthetic colorants in meat products and its application. A multi-level porous ZIF-8 precipitator is used for purification, and a high-performance liquid chromatography (HPLC) method is established for the simultaneous determination of azo synthetic colorants in meat products. Results show that under optimized conditions, the linearity of azo synthetic colorants is good, with a correlation coefficient greater than 0.999, a limit of quantitation (LOQs) of 0.3–0.5 mg / kg, a recovery rate of 92.0%–105.1%, and a relative standard deviation of 0.5%–2.2%. This method can simultaneously detect azo synthetic colorants in meat products, and the multi-level porous ZIF-8 precipitator can be recycled at least 5 times, reducing analytical costs.

[0004] The aforementioned patents improve the detection accuracy of azo synthetic colorants in meat products. However, the methods for detecting meat products only target azo synthetic colorants. In practical applications, the overall detection of meat products is affected by environmental factors and rancidity, making the detection of meat products insufficient and thus reducing the accuracy of meat product food safety detection. Summary of the Invention

[0005] The purpose of this invention is to address the problem that the above-mentioned methods for detecting meat products only target azo synthetic colorants. However, in practical applications, the overall detection of meat products is affected by environmental factors and rancidity, which makes the detection of meat products insufficient and reduces the accuracy of meat product food safety detection. Therefore, this invention proposes a multi-level detection system and method for beef product processing.

[0006] In a first aspect of this invention, a multi-stage detection method for beef product processing is first proposed, the method comprising: The target gas for the target beef product is obtained using a miniature air pump; The target gas is drawn into the common chamber of the dual sensors; the dual sensors consist of two gas sensor units with different functions. The first test of the gas in the common chamber was conducted to obtain the environmental dryness coefficient value; If the ambient dryness coefficient value is less than or equal to the preset dryness coefficient value, the terminal will send a humidity signal. If the environmental dryness coefficient value is greater than the preset dryness coefficient value, the target gas is detected a second time to obtain a gas acidity spectrum. The gas acidity spectrum is then substituted into a neural network model for identification to obtain the probability of an acidic sample. If the probability of the acidic sample is less than or equal to the preset acidity threshold, then image data of the target beef product is acquired, and a third detection is performed on the image data to obtain the meat color result; the meat color result is the position and size of the dark red on the surface of the beef product; if the probability of the acidic sample is greater than the preset acidity threshold, then the terminal sends a spoilage signal.

[0007] Optionally, the principle process of the first detection includes: Within the target time period, obtain the air pressure and airflow rate of the common chamber; calculate the environmental dryness coefficient based on the air pressure and airflow rate.

[0008] The formula for calculating the environmental dryness coefficient is as follows: in, p represents the environmental dryness coefficient value. i V represents the air pressure value at time i, P represents the minimum air pressure value in the shared chamber, and v i Let V represent the airflow rate at time i, V represent the minimum airflow rate in the common chamber, and T represent the target total time.

[0009] Optionally, the second detection specifically involves: detecting the aldehyde content using a specific sensor within a preset time period; and constructing a gas acidity spectrum based on the aldehyde content at each time point. The neural network model specifically includes: The gas acid spectrum is subjected to feature transformation to obtain a first feature map; The first feature map is convolved to obtain the second feature map; The second feature map is discarded to obtain the third feature map; The third feature map is normalized to obtain the output feature map; the output feature map is used to determine the probability of an acidic sample.

[0010] Optionally, the principle process of the third detection includes: The image data is substituted into the image recognition model to obtain the recognition result; the image recognition model is an improvement based on the YOLO8v model, specifically the following improvements: All Conv modules in the backbone and neck networks are replaced with Conv_two modules; The Conv_two module works as follows: The features input to the Conv_two module are used as the input feature tensor; the number of channels of the input feature tensor is C. The input feature tensor is segmented to obtain a first feature tensor with C / 2 channels and a second feature tensor with C / 2 channels; After max pooling the first feature tensor, convolution is performed to obtain a third feature tensor with C / 2 channels; The second feature tensor is convolved twice to obtain a fourth feature tensor with C / 2 channels; The third and fourth feature tensors are concatenated to obtain an output feature tensor with C channels.

[0011] In a second aspect of the invention, a multi-stage detection system for beef product processing is provided, the system comprising: Data acquisition module: Acquires target gas from the target beef product using a miniature air pump; Gas extraction module: extracts the target gas into the common chamber of the dual sensors; the dual sensors consist of two gas sensor units with different functions; First detection module: Performs the first detection of the gas in the common chamber to obtain the environmental dryness coefficient value; The second detection module: If the environmental dryness coefficient value is less than or equal to the preset dryness coefficient value, a humidity signal is sent from the terminal; if the environmental dryness coefficient value is greater than the preset dryness coefficient value, the target gas is detected a second time to obtain a gas acidity spectrum, and the gas acidity spectrum is substituted into a neural network model for identification to obtain the probability of an acidic sample; The third detection module: If the probability of the acidic sample is less than or equal to the preset acidic threshold, then the image data of the target beef product is acquired, and the image data is subjected to a third detection to obtain the meat color result; the meat color result is the position and size of the dark red on the surface of the beef product; if the probability of the acidic sample is greater than the preset acidic threshold, then the terminal sends a sour signal.

[0012] Optionally, the first detection module is further configured to: Within the target time period, obtain the air pressure and airflow rate of the common chamber; calculate the environmental dryness coefficient based on the air pressure and airflow rate.

[0013] The formula for calculating the environmental dryness coefficient is as follows: in, p represents the environmental dryness coefficient value. i V represents the air pressure value at time i, P represents the minimum air pressure value in the shared chamber, and v i V represents the airflow rate at time i, V represents the minimum airflow rate in the common chamber, and T represents the target total time.

[0014] Optionally, the second detection module includes: a detection module and a neural network module. The detection module is used to detect the content of aldehydes through a specific sensor within a preset time; and to form a gas acidity spectrum based on the content of aldehydes at each time point. The neural network module is also used in neural network models, specifically including: The gas acid spectrum is subjected to feature transformation to obtain a first feature map; The first feature map is convolved to obtain the second feature map; The second feature map is discarded to obtain the third feature map; The third feature map is normalized to obtain the output feature map; the output feature map is used to determine the probability of an acidic sample.

[0015] Optionally, the third detection module is further configured to: The image data is substituted into the image recognition model to obtain the recognition result; the image recognition model is an improvement based on the YOLO8v model, specifically the following improvements: All Conv modules in the backbone and neck networks are replaced with Conv_two modules; The Conv_two module works as follows: The features input to the Conv_two module are used as the input feature tensor; the number of channels of the input feature tensor is C. The input feature tensor is segmented to obtain a first feature tensor with C / 2 channels and a second feature tensor with C / 2 channels; After max pooling the first feature tensor, convolution is performed to obtain a third feature tensor with C / 2 channels; The second feature tensor is convolved twice to obtain a fourth feature tensor with C / 2 channels; The third and fourth feature tensors are concatenated to obtain an output feature tensor with C channels.

[0016] Optionally, during model training, the loss function Wise-PLoU is used to replace the original YOLOv8 model's loss function CIoU. The calculation process of the Wise-PLoU loss function includes: in, Indicates the loss value. W represents the squared Euclidean distance of the center offset vector. c H represents the width of the smallest bounding rectangle that encloses the predicted bounding box and the ground truth bounding box. c The height of the smallest bounding rectangle that encloses the predicted bounding box and the ground truth bounding box is represented by , IoU is the intersection-union ratio of the areas of the ground truth bounding box and the predicted bounding box, and e represents a constant.

[0017] The beneficial effects of this invention are: This invention proposes a multi-stage detection system and method for beef product processing. A rapid initial detection determines environmental dryness, preemptively eliminating interference from humidity. Only after the environment has reached the required dryness level is a secondary gas acidity spectrum detection performed, utilizing a neural network model to intelligently identify the probability of rancidity, significantly improving detection accuracy. Subsequently, based on the identification results, an adaptive decision is made to initiate a third meat color image detection, achieving joint judgment of gas and visual characteristics. This solution can accurately locate deep red spoiled areas on the surface of beef while reducing unnecessary detection steps, effectively improving the accuracy of beef product quality detection and reducing the false detection rate. Attached Figure Description

[0018] The invention will now be further described with reference to the accompanying drawings.

[0019] Figure 1 A flowchart of a multi-stage detection method for beef product processing provided in an embodiment of the present invention; Figure 2 A network structure diagram of a YOLO8v model provided in an embodiment of the present invention; Figure 3 This is a network structure diagram of an image recognition model provided in an embodiment of the present invention; Figure 4 This is a framework diagram of a multi-level detection system for beef product processing provided in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and B can represent: A alone, A and B simultaneously, and B alone. Furthermore, descriptions involving "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" can explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0021] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] This invention provides a multi-stage detection method for beef product processing. See also... Figure 1 , Figure 1 A flowchart illustrating a multi-stage detection method for beef product processing provided in an embodiment of the present invention. The method includes the following steps: The target gas for the target beef product is obtained by a miniature air pump; the target gas is then drawn into the common chamber of the dual sensors; the dual sensors consist of two gas sensor units with different functions. The first test of the gas in the common chamber was conducted to obtain the environmental dryness coefficient value; If the ambient dryness coefficient value is less than or equal to the preset dryness coefficient value, the terminal will send a humidity signal. If the environmental dryness coefficient value is greater than the preset dryness coefficient value, the target gas is detected a second time to obtain a gas acidity spectrum. The gas acidity spectrum is then substituted into the neural network model for identification to obtain the probability of an acidic sample. If the probability of acidic samples is less than or equal to the preset acidity threshold, the image data of the target beef product is acquired, and a third detection is performed on the image data to obtain the meat color result; the meat color result is the position and size of the dark red on the surface of the beef product; if the probability of acidic samples is greater than the preset acidity threshold, the terminal sends a sour signal.

[0023] Based on the multi-stage detection method for beef product processing provided by this invention, the first detection quickly determines the dryness of the environment, eliminating moisture interference in advance; the second gas acid spectrum detection is only performed after the drying conditions meet the standards, and the probability of rancidity is intelligently identified by a neural network model, which greatly improves the detection accuracy; then, based on the identification results, an adaptive decision is made to initiate a third detection of the meat color image, realizing the joint judgment of gas features and visual features. This method can accurately locate the dark red spoiled area on the surface of beef, reduce invalid detection processes, effectively improve the accuracy of beef quality detection, and reduce false detections.

[0024] Specifically, the humidity signal indicates that the air around the beef product is humid, and the spoilage signal indicates that the beef product has become acidic; the common chamber is a shared area for detection by the dual sensors; the preset acidity threshold and preset dryness coefficient values ​​are calculated by staff based on historical parameters and experience.

[0025] In one implementation method, the principle process of the first detection includes: Within the target time period, obtain the air pressure and airflow rate of the common chamber; calculate the environmental dryness coefficient based on the air pressure and airflow rate.

[0026] The formula for calculating the environmental aridity coefficient is: in, p represents the environmental dryness coefficient value. i V represents the air pressure value at time i, P represents the minimum air pressure value in the shared chamber, and v i Let V represent the airflow rate at time i, V represent the minimum airflow rate in the common chamber, and T represent the target total time.

[0027] In one implementation, a higher air pressure value indicates a higher density of water molecules in the common chamber, resulting in higher humidity and a lower environmental dryness coefficient. A higher airflow rate indicates that the high-speed airflow carries away more water molecules, resulting in a lower environmental dryness coefficient. The environmental dryness coefficient is obtained by evaluating the dryness of the common chamber using air pressure and airflow rate. The target time period can be a preset fixed duration interval (such as 1s, 5s, 10s, 1min, etc., configured by the system).

[0028] In one implementation, the second detection specifically includes: detecting the aldehyde content using a specific sensor within a preset time period; and constructing a gas acidity spectrum based on the aldehyde content at each time point. Neural network models specifically include: The first feature map is obtained by performing feature transformation on the gas acid spectrum; The second feature map is obtained by convolving the first feature map; The third feature map is obtained by discarding features from the second feature map; The third feature map is normalized to obtain the output feature map; the output feature map is used to determine the probability of an acidic sample.

[0029] In one implementation, the computation process of the neural network model is as follows: Specifically, T represents the output feature map calculated by the neural network model, Conv represents feature convolution, Dropout represents feature discarding, ReLU() represents the activation function, which speeds up the training process and enhances the model's feature extraction capability. Pooling layers are used to reduce the feature space after convolution and reduce the number of network parameters; E represents the first feature map for feature transformation.

[0030] In one implementation, the principle and process of the third detection include: See Figure 2 , Figure 2 This is a network structure diagram of a YOLO8v model provided in an embodiment of the present invention, in which image data is substituted into the image recognition model to obtain the recognition result; see also Figure 3 , Figure 3 This is a network structure diagram of an image recognition model provided in an embodiment of the present invention. The image recognition model is an improvement based on the YOLO8v model, specifically as follows: All Conv modules in the backbone and neck networks are replaced with Conv_two modules; The Conv_two module works as follows: The features input to the Conv_two module are used as the input feature tensor; the number of channels of the input feature tensor is C; and C is a value greater than zero. The input feature tensor is segmented to obtain a first feature tensor with C / 2 channels and a second feature tensor with C / 2 channels; After max pooling the first feature tensor, convolution is performed to obtain a third feature tensor with C / 2 channels; The second feature tensor is convolved twice to obtain a fourth feature tensor with C / 2 channels; The third and fourth feature tensors are concatenated to obtain an output feature tensor with C channels.

[0031] Specifically, the YOLO8v model's network structure includes a 22-layer basic structure and three detection heads; In one implementation, the input feature map has C channels, processed through two complementary branches, with the channels split into a fixed 50%:50% ratio. These branches are a pooling branch (using MaxPool2d for spatial downsampling with a 2×2 kernel and a stride of 2, followed by a 1×1 convolution to adjust the output channels to C / 2; this branch focuses on extracting low-frequency, globally salient features while preserving edges) and a convolution branch (adjusting the channels through a 1×1 convolution, simplifying the input channels to a 3×3 convolution with a stride of 2 for spatial downsampling, while preserving local structure and texture details). The outputs of the two branches are concatenated to generate the final feature map, balancing shallow details and deep semantic information. This approach enhances the preservation and representation of small object features while controlling computational costs; it allows adjustment of input and output channels while maintaining a fixed channel split ratio, improving the accuracy of detecting the deep red surface color of beef products.

[0032] In one implementation, during model training, the loss function Wise-PLoU is used to replace the original YOLOv8 model's loss function CIoU. The calculation process of the loss function Wise-PLoU includes: in, Indicates the loss value. W represents the squared Euclidean distance of the center offset vector. c H represents the width of the smallest bounding rectangle that encloses the predicted bounding box and the ground truth bounding box. c The height of the smallest bounding rectangle that encloses the predicted bounding box and the ground truth bounding box is represented by , IoU is the intersection-union ratio of the areas of the ground truth bounding box and the predicted bounding box, and e represents a constant.

[0033] The multi-level detection method for beef product processing provided in this invention utilizes the YOLOv8 model to learn complex texture features from a large number of samples, effectively distinguishing normal patterns from dark red patterns and significantly improving recognition accuracy. Simultaneously, the addition of a small target detection layer enhances the model's sensitivity in detecting small target defects, significantly improving its ability to detect small targets such as dark red patterns and reducing missed detections.

[0034] Based on the same inventive concept, this invention also provides a multi-level detection system for beef product processing. See [link to related document]. Figure 4 , Figure 4 A framework diagram of a multi-level detection system for beef product processing provided in an embodiment of the present invention includes: Data acquisition module: Acquires target gas from the target beef product using a miniature air pump; Gas extraction module: Extracts the target gas into the common chamber of the dual sensors; the dual sensors consist of two gas sensor units with different functions. First detection module: Performs the first detection of the gas in the common chamber to obtain the environmental dryness coefficient value; The second detection module: If the environmental dryness coefficient value is less than or equal to the preset dryness coefficient value, the terminal sends a humidity signal; if the environmental dryness coefficient value is greater than the preset dryness coefficient value, the target gas is detected a second time to obtain a gas acidity spectrum, and the gas acidity spectrum is substituted into the neural network model for identification to obtain the probability of acidic sample. The third detection module: If the probability of acidic samples is less than or equal to the preset acid threshold, the image data of the target beef product is acquired, and the image data is subjected to a third detection to obtain the meat color result; the meat color result is the position and size of the dark red on the surface of the beef product; if the probability of acidic samples is greater than the preset acid threshold, the terminal sends a sour signal.

[0035] Based on the multi-level detection system for beef product processing provided in this invention, the first detection quickly determines the dryness of the environment, eliminating moisture interference in advance; a second gas acid spectrum detection is performed only after the drying conditions meet the standards, and a neural network model is used to intelligently identify the probability of rancidity, significantly improving detection accuracy; then, based on the identification results, an adaptive decision is made to initiate a third detection of the meat color image, realizing the joint judgment of gas features and visual features. This can accurately locate the dark red spoiled areas on the surface of the beef, reduce invalid detection processes, effectively improve the accuracy of beef quality detection, and reduce false detections.

[0036] The foregoing has described one embodiment of the present invention in detail, but this content is merely a preferred embodiment and should not be considered as limiting the scope of the present invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims of this invention.

Claims

1. A multi-stage detection method for beef product processing, characterized in that, The method includes: The target gas for the target beef product is obtained using a miniature air pump; The target gas is drawn into the common chamber of the dual sensors; the dual sensors consist of two gas sensor units with different functions. The first test of the gas in the common chamber was conducted to obtain the environmental dryness coefficient value; If the ambient dryness coefficient value is less than or equal to the preset dryness coefficient value, the terminal will send a humidity signal. If the environmental dryness coefficient value is greater than the preset dryness coefficient value, the target gas is detected a second time to obtain a gas acidity spectrum. The gas acidity spectrum is then substituted into a neural network model for identification to obtain the probability of an acidic sample. If the probability of the acidic sample is less than or equal to the preset acidity threshold, then image data of the target beef product is acquired, and a third detection is performed on the image data to obtain the meat color result; the meat color result is the position and size of the dark red on the surface of the beef product; if the probability of the acidic sample is greater than the preset acidity threshold, then the terminal sends a spoilage signal.

2. The multi-stage detection method for beef product processing according to claim 1, characterized in that, The principle and process of the first detection include: Within the target time period, the air pressure and airflow rate of the common chamber are acquired; the environmental dryness coefficient is calculated based on the air pressure and airflow rate. The formula for calculating the environmental dryness coefficient is as follows: in, p represents the environmental dryness coefficient value. i V represents the air pressure value at time i, P represents the minimum air pressure value in the shared chamber, and v i Let V represent the airflow rate at time i, V represent the minimum airflow rate in the common chamber, and T represent the target total time.

3. The multi-stage detection method for beef product processing according to claim 1, characterized in that, The second detection specifically includes: detecting the aldehyde content using a specific sensor within a preset time period; and constructing a gas acidity spectrum based on the aldehyde content at each time point. The principle process of the neural network model includes: The gas acid spectrum is subjected to feature transformation to obtain a first feature map; The first feature map is convolved to obtain the second feature map; The second feature map is discarded to obtain the third feature map; The third feature map is normalized to obtain the output feature map; the output feature map is used to determine the probability of an acidic sample.

4. The multi-stage detection method for beef product processing according to claim 1, characterized in that, The principle and process of the third detection include: The image data is substituted into the image recognition model to obtain the recognition result; the image recognition model is an improvement based on the YOLO8v model, specifically the following improvements: All Conv modules in the backbone and neck networks are replaced with Conv_two modules; The Conv_two module works as follows: The features input to the Conv_two module are used as the input feature tensor; the number of channels of the input feature tensor is C. The input feature tensor is segmented to obtain a first feature tensor with C / 2 channels and a second feature tensor with C / 2 channels; After max pooling the first feature tensor, convolution is performed to obtain a third feature tensor with C / 2 channels; The second feature tensor is convolved twice to obtain a fourth feature tensor with C / 2 channels; The third and fourth feature tensors are concatenated to obtain an output feature tensor with C channels.

5. The multi-stage detection method for beef product processing according to claim 4, characterized in that, During model training, the loss function Wise-PLoU is used to replace the original YOLOv8 model's loss function CIoU. The calculation process of the Wise-PLoU loss function includes: in, Indicates the loss value. W represents the squared Euclidean distance of the center offset vector. c H represents the width of the smallest bounding rectangle that encloses the predicted bounding box and the ground truth bounding box. c The height of the smallest bounding rectangle that encloses the predicted bounding box and the ground truth bounding box is represented by , IoU is the intersection-union ratio of the areas of the ground truth bounding box and the predicted bounding box, and e represents a constant.

6. A multi-stage detection system for beef product processing, characterized in that, The system includes: Data acquisition module: Acquires target gas from the target beef product using a miniature air pump; Gas extraction module: extracts the target gas into the common chamber of the dual sensors; the dual sensors consist of two gas sensor units with different functions; First detection module: Performs the first detection of the gas in the common chamber to obtain the environmental dryness coefficient value; The second detection module: If the environmental dryness coefficient value is less than or equal to the preset dryness coefficient value, a humidity signal is sent from the terminal; if the environmental dryness coefficient value is greater than the preset dryness coefficient value, the target gas is detected a second time to obtain a gas acidity spectrum, and the gas acidity spectrum is substituted into a neural network model for identification to obtain the probability of an acidic sample; The third detection module: If the probability of the acidic sample is less than or equal to the preset acidic threshold, then the image data of the target beef product is acquired, and the image data is subjected to a third detection to obtain the meat color result; the meat color result is the position and size of the dark red on the surface of the beef product; if the probability of the acidic sample is greater than the preset acidic threshold, then the terminal sends a sour signal.

7. The multi-stage detection system for beef product processing according to claim 6, characterized in that, The first detection module is also used for: Within the target time period, acquire the air pressure and airflow rate of the common chamber; calculate the environmental dryness coefficient based on the air pressure and airflow rate; The formula for calculating the environmental dryness coefficient is as follows: in, p represents the environmental dryness coefficient value. i V represents the air pressure value at time i, P represents the minimum air pressure value in the shared chamber, and v i Let V represent the airflow rate at time i, V represent the minimum airflow rate in the common chamber, and T represent the target total time.

8. The multi-stage detection system for beef product processing according to claim 6, characterized in that, The second detection module includes: a detection module and a neural network module. The detection module is used to detect the content of aldehydes through a specific sensor within a preset time; and to form a gas acidity spectrum based on the content of aldehydes at each time point. The neural network module is also used in neural network models, specifically including: The gas acid spectrum is subjected to feature transformation to obtain a first feature map; The first feature map is convolved to obtain the second feature map; The second feature map is discarded to obtain the third feature map; The third feature map is normalized to obtain the output feature map; the output feature map is used to determine the probability of an acidic sample.

9. A multi-stage detection system for beef product processing according to claim 6, characterized in that, The third detection module is also used for: The image data is substituted into the image recognition model to obtain the recognition result; the image recognition model is an improvement based on the YOLO8v model, specifically the following improvements: All Conv modules in the backbone and neck networks are replaced with Conv_two modules; The Conv_two module works as follows: The features input to the Conv_two module are used as the input feature tensor; the number of channels of the input feature tensor is C. The input feature tensor is segmented to obtain a first feature tensor with C / 2 channels and a second feature tensor with C / 2 channels; After max pooling the first feature tensor, convolution is performed to obtain a third feature tensor with C / 2 channels; The second feature tensor is convolved twice to obtain a fourth feature tensor with C / 2 channels; The third and fourth feature tensors are concatenated to obtain an output feature tensor with C channels.

10. A multi-stage detection system for beef product processing according to claim 9, characterized in that, During model training, the loss function Wise-PLoU is used to replace the original YOLOv8 model's loss function CIoU. The calculation process of the Wise-PLoU loss function includes: in, Indicates the loss value. W represents the squared Euclidean distance of the center offset vector. c H represents the width of the smallest bounding rectangle that encloses the predicted bounding box and the ground truth bounding box. c The height of the smallest bounding rectangle that encloses the predicted bounding box and the ground truth bounding box is represented by , IoU is the intersection-union ratio of the areas of the ground truth bounding box and the predicted bounding box, and e represents a constant.

Images