Raw chicken carcass damage grading detection device and method based on fusion of image and smell sensor

Abstract

The invention discloses a raw chicken carcass damage grading detection device based on fusion of an image and a smell sensor. The raw chicken carcass damage grading detection device comprises a transmission unit, an image acquisition unit, a smell detection unit, a data processing unit and a grading output unit. Meanwhile, the invention discloses a method for detecting by using the raw chicken carcass damage grading detection device based on fusion of the image and the smell sensor, synchronous and accurate grading of the raw chicken carcass damage and freshness is realized through multi-dimensional data fusion, and reliable technical support is provided for screening of chicken pretreatment raw materials.

Description

Technical Field This invention relates to the field of food testing technology, specifically to a device and method for detecting damage grading in raw chicken carcasses based on the fusion of image and odor sensors, applicable to the screening of raw chicken carcasses in the pre-processing stage of braised chicken production. Background Technology In the braised chicken processing industry chain, the damage (such as skin damage and subcutaneous hematoma) and freshness of the raw chicken carcass directly determine the quality of subsequent products. Among existing damage detection technologies, image recognition technology can accurately locate surface damage, but it cannot determine whether the damaged area is accompanied by spoilage caused by microbial growth; odor sensor technology can detect the freshness of the carcass, but it is difficult to locate the location of damage and quantify the degree of damage. Using the aforementioned technologies alone has significant drawbacks: relying solely on image recognition may confuse damaged but not spoiled carcasses with damaged and spoiled ones, leading to waste of qualified raw materials or the influx of substandard materials into production; relying solely on odor sensors cannot distinguish between undamaged but spoiled carcasses and damaged and spoiled ones, nor can it provide location guidance for subsequent damage repair (such as hematoma removal). Therefore, there is an urgent need for a technology that integrates image and odor detection to achieve integrated detection of damage location, severity grading, and freshness assessment. Summary of the Invention This invention aims to overcome the shortcomings of existing technologies and provide a device and method for grading and detecting damage in raw chicken carcasses based on the fusion of image and odor sensors. Through multi-dimensional data fusion, it achieves simultaneous and accurate grading of damage and freshness in raw chicken carcasses, providing reliable technical support for the screening of raw materials for braised chicken pretreatment. A damage grading detection device for raw chicken carcasses based on image and odor sensor fusion includes a transmission unit, an image acquisition unit, an odor detection unit, a data processing unit, and a grading output unit. The connection relationships of each unit are as follows: Conveying unit: includes stainless steel conveyor belt, drive motor and positioning fixtures. The positioning fixtures are spaced 30-40cm apart and are used to fix the raw chicken carcasses (breast facing up, uniform posture). The conveyor belt speed is adjustable (0.5-1m / min) to adapt to different testing efficiency requirements. Any existing technology that can achieve positioning, clamping and conveying functions can be used. Image acquisition unit: Located 50-60cm above the conveyor unit, it includes two industrial cameras (24-megapixel resolution, 30fps frame rate), four ring LED fill lights (5500K color temperature, 800 lux illuminance) and a background board (white matte material to avoid glare); one camera acquires the frontal image of the body and the other acquires the side image, achieving full-angle coverage. Odor detection unit: includes odor acquisition probes, gas sensor array, and gas pretreatment module; the odor acquisition probes are made of stainless steel with an aperture of 0.5mm, and are located on both sides of the transmission unit (3-5cm from the carcass surface), with a total of 4 probes; the gas sensor array contains 6 sensors, corresponding to ammonia, hydrogen sulfide, trimethylamine, and other gases characterized by damage and deterioration; the gas pretreatment module includes a dehumidifier (dew point controlled below -40℃) and a filter (filtration accuracy of 0.1μm) to avoid interference from water vapor and impurities in the detection. Data processing unit: An embedded industrial computer with built-in image processing algorithm module and odor data analysis module, which can realize multi-source data fusion calculation. The grading output unit includes a display screen (which displays the test results in real time), an audible and visual alarm (which triggers an alarm when a carcass is found to be defective), and a sorting pusher (which pushes the carcass to the corresponding discharge port according to the grading results). Any existing device that can achieve this function can be used. The method for grading damage in raw chicken carcasses based on the above-mentioned device includes the following steps: (1) Sample initialization and parameter calibration Place the standard sample (a raw chicken carcass with known damage level and freshness) into the positioning fixture and start the device; the image acquisition unit acquires images of the standard sample and establishes a damage type-image feature database (such as the edge gray value of skin damage and the RGB color gamut range of bruised areas); the odor detection unit acquires the characteristic gas concentration of the standard sample and sets freshness thresholds (ammonia concentration <0.3ppm and hydrogen sulfide concentration <0.05ppm are considered fresh, and exceeding these thresholds indicates spoilage). (2) Live chicken carcass transport and multi-source data acquisition The raw chicken carcass to be tested is fixed in the positioning fixture, and the conveyor belt transports it to the testing area; the image acquisition unit simultaneously acquires front and side images (acquisition time < 0.5s), the odor acquisition probe absorbs the gas on the surface of the carcass, and after preprocessing, it is sent to the gas-sensitive sensor array to acquire characteristic gas concentration data (response time < 1s).

[0001] A gas-sensitive sensor array (containing three types of sensors: ammonia, hydrogen sulfide, and trimethylamine) is used to collect characteristic gas concentration data (response time < 1s, detection accuracy 0.01ppm). (3) Image data processing and damage identification The image processing algorithm module preprocesses the acquired images: first, it removes noise by Gaussian filtering (filter kernel size 5×5), then it extracts the body region by adaptive threshold segmentation, and finally it identifies the damage type (skin damage / subcutaneous hematoma), locates the damage location (accuracy ±1mm) and calculates the damage area by Canny edge detection and RGB color gamut analysis. 3.1 Hyperspectral Image Correction: Systematic errors are eliminated using a black-and-white correction method. The correction equation is as follows: ; Where I is the corrected image, I0 is the original hyperspectral image, W is the reference image of a standard white board (99% reflectivity) under the same shooting conditions, and B is the dark reference image when the camera lens is closed (used to remove the influence of sensor dark current). (4) Odor data processing and freshness assessment The odor data analysis module normalizes the concentration data collected by the sensor array (converting the concentration values ​​into standardized values ​​of 0-1) and compares them with preset thresholds: if the concentration of all characteristic gases is below the threshold, it is determined to be fresh; if the concentration of any characteristic gas exceeds the threshold, it is determined to be spoiled, and the type of gas exceeding the standard is recorded (e.g., excessive ammonia corresponds to protein putrefaction). Gas data normalization: The concentration data collected by the gas sensor is converted into a 0-1 standardized value. The normalization formula is as follows: ; Among them, C raw C represents the original concentration value. min The detection limits for this gas are (ammonia 0.05 ppm, hydrogen sulfide 0.01 ppm, trimethylamine 0.02 ppm), C max The upper limit of the safety threshold (ammonia 0.8ppm, hydrogen sulfide 0.2ppm, trimethylamine 0.5ppm).

[0002] (5) Multi-source data fusion and hierarchical output The data processing unit integrates the damage identification results with the freshness assessment results and outputs a grading system according to the following criteria: Grade 1 (Premium): Undamaged and fresh, the sorting pusher pushes the carcass to the direct pre-processing discharge port; Level 2 (Repairable): Minor damage (damaged area < 1 cm² or bruised area < 3 cm²) + fresh, output the coordinates of the damaged location and push it to the repair processing outlet; Level 3 (Restricted Use): Moderate damage (damaged area 1-3cm² or bruised area 3-10cm²) + fresh, push to the local excision outlet; Level 4 (Unqualified): Severe damage (damaged area > 3cm² or bruising area > 10cm²) or any level of damage + deterioration, triggering an audible and visual alarm, and pushing to the waste discharge port. Beneficial effects This invention, through the fusion of image and odor sensors, achieves for the first time the simultaneous detection of damage to raw chicken carcasses and the assessment of freshness, solving the problem that a single technology cannot comprehensively evaluate the quality of raw materials. The grading accuracy rate reaches over 96%, which is a significant improvement over traditional single technologies. The image processing module can accurately locate the damage location and quantify the damage area, providing clear guidance for subsequent repair procedures (such as hematoma removal), reducing human judgment errors, and increasing raw material utilization by more than 15%; the odor detection module can identify the risk of spoilage in advance, avoiding food safety issues caused by microbial contamination. The device is highly automated, with a total inspection time of less than 2 seconds per carcass. It can be adapted to high-speed production lines with a capacity of 600 chickens / hour or more. The inspection process is non-contact and non-damaging, and will not cause secondary contamination to the raw chicken carcasses, thus meeting food processing hygiene standards. Attached Figure Description Figure 1 This is a flowchart of the apparatus of the present invention. Detailed Implementation The following embodiments are merely examples illustrating implementations of the present invention and do not constitute any limitation on the present invention. Those skilled in the art will understand that modifications made without departing from the spirit and concept of the present invention fall within the protection scope of the present invention. Unless otherwise specified, the reagents and instruments used in the following embodiments are commercially available products.

[0003] Equipment configuration: hyperspectral imaging system (resolution 1632×1024 pixels), gas sensor array (response time <0.8s), automated conveyor belt (positioning accuracy ±0.5mm), industrial computer (CPU i7-12700K, memory 32GB); Sample size: 40 live chicken carcasses (weighing 1.5-2.0 kg) were selected, and different types of injuries were artificially created (skin lesions: 0.5-5 cm). 2 A total of 80 areas; subcutaneous bruising: 1-12cm 2 A total of 100 areas were set up, and samples of different freshness were set up (fresh: within 2 hours after slaughter; slightly spoiled: stored at 25℃ for 12 hours; spoiled: stored at 25℃ for 24 hours).

[0004] Example 1 (1) Establish a damage type-image feature database and set a freshness threshold for characteristic gases; (ammonia concentration < 0.3 ppm and hydrogen sulfide concentration < 0.05 ppm are considered fresh, and those exceeding this threshold are considered deteriorated). (2) Sample preparation and multi-source data acquisition 2.1 Sample fixation and environmental control: The raw chicken carcasses to be tested are fixed with the breast and abdomen facing upwards in a positioning fixture covered with black light-absorbing cloth (fixture error ±0.5mm), and placed on an automated conveyor belt (conveyor belt speed 0.5-1m / min). The ambient temperature is controlled at 20-25℃ and the relative humidity at 50-60%, and external light interference is avoided. 2.2 Hyperspectral Image Acquisition: One hyperspectral imaging system was set up above the detection area (350-60cm from the carcass surface) and on the side (horizontal angle 45°). The camera exposure time was 350-400ms, and the acquisition band range was 380-1050nm (including 1232 continuous bands). Hyperspectral images of the front and side of the carcass were acquired simultaneously (acquisition time for a single carcass <0.5s) and stored in RAW format. 2.3 Characteristic Gas Collection: Two odor collection probes (0.5mm aperture) are installed 5-8cm downstream of the hyperspectral collection area, 3-5cm from the carcass surface. A vacuum pump (0.02MPa negative pressure) is used to draw volatile gases from the carcass surface. After pretreatment in a silica gel drying tube (dehumidification efficiency >95%), the gases are fed into a gas-sensitive sensor array (containing sensors for ammonia, hydrogen sulfide, and trimethylamine) to collect characteristic gas concentration data (response time <1s, detection accuracy 0.01ppm). Gas pretreatment of the carcass can prevent interference from moisture and impurities.

[0005] (3) Data preprocessing 3.1 Hyperspectral Image Correction: Systematic errors are eliminated using a black-and-white correction method. The correction equation is as follows: ; Where I is the corrected image, I0 is the original hyperspectral image, W is the reference image of a standard white board (99% reflectivity) under the same shooting conditions, and B is the dark reference image when the camera lens is closed (used to remove the influence of sensor dark current). 3.2 Gas Data Normalization: The concentration data collected by the gas sensor is converted into 0-1 standardized values. The normalization formula is as follows: ; Among them, C raw C represents the original concentration value. min The detection limits for this gas are (ammonia 0.05 ppm, hydrogen sulfide 0.01 ppm, trimethylamine 0.02 ppm), C max The upper limit of the safety threshold (ammonia 0.8ppm, hydrogen sulfide 0.2ppm, trimethylamine 0.5ppm).

[0006] (4) Hyperspectral image damage identification and quantification 4.1 Region of Interest (ROI) Extraction: The corrected hyperspectral image was imported using ENVI 5.3 software, and three types of regions were selected using the ROI tool: Damage ROIs: These include areas of skin damage (edge ​​grayscale value 100-150) and subcutaneous bruising (RGB color gamut R>180, G<120, B<100). Edge-mixed pixels were avoided during selection. A total of 280 damage ROIs were collected (120 for skin damage and 160 for bruising). Normal ROI: 35 normal areas, including the breast, drumstick, and wing of the carcass, were extracted using linear ROIs. Background ROI: Extract the black fixture regions, a total of 35; 4.2 Two-dimensional correlation spectrum construction and feature band extraction: 4.2.1 The 280 damaged ROI data are constructed into m damage sets (m=280), which are considered as perturbations to the clean carcass. The reflectance of the j-th set in the λnm band is... (t) j (where j is the j-th perturbation variable), the dynamic spectrum is represented as: ; Where A0 is the average spectrum (reference spectrum) of a normal ROI; 4.2.2 Calculate the synchronization correlation strength: ; Where λa and λb are two randomly selected bands, the calculation results are presented as a two-dimensional contour map, and the hyperspectral synchronous two-dimensional correlation spectrum is obtained. 4.2.3 Feature band selection: Extract the bands with autocorrelation peak intensity > 0.8 in the two-dimensional correlation spectrum and determine the damage identification feature bands as 468nm (sensitive to bruising), 652nm (sensitive to damage), and 820nm (background differentiation). 4.3 False-color image enhancement and impairment quantization: 4.3.1 False Color Fusion: Input 652nm band data into the R channel, 468nm into the G channel, and 820nm into the B channel. Generate a false color image using the ENVI band fusion function. At this time, the damaged skin area is red (R>0.6, G<0.4, B<0.3), the subcutaneous bruising area is orange (R>0.5, G>0.3, B<0.2), and normal skin is light blue (R<0.3, G>0.5, B>0.6). 4.3.2 Damage area calculation: The false-color image was converted into a double-precision image using the MATLAB "im2double" function. The contour of the damaged area was extracted by Canny edge detection (threshold 100-200), and the pixel area (1 pixel corresponds to 0.01cm²) was calculated with an accuracy of ±0.05cm². 4.4 Threshold Optimization and False Positive Removal: The ROC curve method was used to determine the segmentation thresholds for the R, G, and B channels. Plot an ROC curve with the damage area recognition accuracy (TPR) as the ordinate and the normal area misclassification rate (FPR) as the abscissa. Iterate through the three-channel threshold combinations (R: 0.4-0.7, G: 0.2-0.5, B: 0.1-0.4) and select the threshold combination with the smallest Euclidean distance to the ideal point (TPR=1, FPR=0) (optimal threshold: R=0.55, G=0.35, B=0.25). False positives were removed by region labeling: connected regions with an area of ​​less than 50 pixels (corresponding to 0.5 cm²) were cleared, reducing the false positive rate to below 3%.

[0007] (5) Characteristic gas analysis and freshness determination 5.1 Freshness Threshold Setting: Based on the spoilage mechanism of raw chicken carcasses, characteristic gas thresholds are set as follows: Fresh: Ammonia concentration < 0.3 ppm, hydrogen sulfide concentration < 0.05 ppm, trimethylamine concentration < 0.1 ppm; Slight spoilage: ammonia concentration 0.3-0.5 ppm, hydrogen sulfide concentration 0.05-0.1 ppm, or trimethylamine concentration 0.1-0.2 ppm; Putrefaction: Ammonia concentration > 0.5 ppm, hydrogen sulfide concentration > 0.1 ppm, or trimethylamine concentration > 0.2 ppm; 4.2 Gas Data Judgment: Compare the normalized gas concentration data with the above thresholds, output the freshness level (fresh / slightly spoiled / spoiled), and record the type of gas exceeding the standard (e.g., excessive ammonia corresponds to protein spoilage, excessive hydrogen sulfide corresponds to the decomposition of sulfur-containing amino acids).

[0008] Step 5: Multi-source data fusion and hierarchical output Based on the results of damage quantification and freshness level, a four-level grading standard was established: 5.2 Results Visualization and Feedback: The carcass grading results, damage location coordinates (marked with colored boxes on the hyperspectral image), and concentration of excessive gases are displayed in real time on an industrial display screen. At the same time, the data is uploaded to the database (storage format: CSV) to support historical data traceability.

[0009] Experimental Example 1 Testing process: The image acquisition unit acquires images, processes them, identifies skin damage, locates the damage on the left side of the chicken breast, and calculates the area to be 0.8 cm². 2 ; The odor detection unit collects gas; if the concentration of characteristic gases is below the threshold, the gas is determined to be fresh. After data fusion, a secondary (repairable) output is generated. The display shows the coordinates of the damaged location (X: 120mm, Y: 80mm), and the sorting pusher pushes it to the repair processing outlet. Experimental Example 2 Testing process: Image processing identifies subcutaneous bruising, area 12cm 2 The injury was determined to be severe. Odor testing showed that the ammonia concentration exceeded the standard, indicating that the product had spoiled. After data fusion, a level 4 (unqualified) output is generated, triggering an audible and visual alarm and pushing the material to the waste discharge port. The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A device for grading and detecting damage in raw chicken carcasses based on the fusion of image and odor sensors, characterized in that, It includes a transmission unit, an image acquisition unit, an odor detection unit, a data processing unit, and a graded output unit connected in sequence; The conveying unit is used to convey raw chicken carcasses; The image acquisition unit is used to acquire images of the carcass and includes two industrial cameras, four ring LED fill lights and a white matte background plate, which are set 50-60cm above the conveying unit; The odor detection unit is used to collect the odor of raw chicken carcasses; it includes an array of gas-sensitive sensors. The gas pretreatment module is used to handle water vapor and impurities. The data processing unit is an embedded industrial computer with a built-in image processing algorithm module and an odor data analysis module to realize multi-source data fusion calculation. The graded output unit is used to grade and output live chickens.

2. The raw chicken carcass damage grading detection device based on image and odor sensor fusion according to claim 1, characterized in that, The gas sensor array detects ammonia, hydrogen sulfide, and organic acids, and the gas pretreatment module controls the dew point to below -40°C.

3. A method for grading and detecting damage in raw chicken carcasses based on the device described in claim 1, characterized in that, Includes the following steps: (1) Sample initialization and parameter calibration: Establish a database of damage types and image features, and set a freshness threshold for characteristic gases; (2) Multi-source data acquisition: The conveyor belt transports the carcass to the detection area, the image acquisition unit acquires images of the front and back of the live chicken, and the odor detection unit acquires the concentration of characteristic gases of the live chicken; (3) Image data processing: Identify the type, location and area of ​​damage through Gaussian filtering, threshold segmentation and edge detection; (4) Odor data processing: Normalize the concentration data and compare it with the threshold to determine freshness; (5) Data fusion and classification: The four-level results are output according to the damage level and freshness, which drive the sorting push rod and alarm device.

4. The detection method according to claim 3, characterized in that, The grading standards in step (5) are: Level 1, Level 2, Level 3, and Level 4.

5. The detection method according to claim 3, characterized in that, The grading criteria in step (5) are as follows: Level 1: No damage, fresh; Level 2: Slight damage, fresh; Level 3: Moderate damage, fresh; Level 4: Severe damage, spoiled.

Images