Fruit and vegetable fresh-keeping method based on subacid electrolyzed water dry fog

Abstract

The invention discloses a fruit and vegetable fresh-keeping method based on subacid electrolyzed water dry fog, and relates to the technical field of fruit and vegetable fresh-keeping, and the fruit and vegetable fresh-keeping method comprises the following steps: atomizing subacid electrolyzed water according to a set initial particle size range to form dry fog, and acting the dry fog on the surfaces of fruits and vegetables in a high-speed assembly line transmission process to form a wide-area sterilization protection fog layer; acquiring multispectral image information of the outer surfaces of the fruits and vegetables through a multispectral image acquisition device when the fruits and vegetables pass through the assembly line; and processing the collected original image, and matching the image frame with the position of the fruit and vegetable on the assembly line according to the assembly line speed and the imaging frequency. The fruit and vegetable crack severity degree is analyzed through multispectral image recognition and the convolutional neural network, the dry fog particle size is regulated in real time, differential fresh-keeping treatment is achieved, the sterilization effect of fog drops attached to the crack area is improved, the problem that traditional dry fog with the fixed particle size is uneven in coverage is effectively solved, and the fresh-keeping precision and the anti-pollution capacity are remarkably improved.

Description

Technical Field

[0001] The invention relates to the technical field of fruit and vegetable preservation, and in particular to a fruit and vegetable preservation method based on slightly acidic electrolyzed water dry mist. Background Art

[0002] Using slightly acidic electrolyzed water dry mist to preserve fruits and vegetables means atomizing electrolyzed water (usually containing low concentrations of hypochlorous acid) with bactericidal and disinfecting effects and in a weakly acidic state into extremely fine dry mist particles, and spraying or soaking the fruits and vegetables in a specific environment. Since the dry mist particles are extremely small and evenly distributed, it can quickly penetrate into the gaps or wrinkles on the surface of fruits and vegetables without increasing the moisture burden on the surface of fruits and vegetables, effectively inhibiting the growth of microorganisms such as bacteria and fungi, thereby slowing down corruption and extending the shelf life. In addition, slightly acidic electrolyzed water is usually relatively safe and mild, which can reduce the impact on the quality and taste of fruits and vegetables, and achieve a more environmentally friendly and efficient preservation effect.

[0003] During the sorting, grading, weighing and other processing processes of fruits and vegetables in high-speed production lines, slightly acidic electrolyzed water dry mist is often used for preservation to form a sterilization protective layer on the surface of fruits and vegetables to inhibit microbial contamination introduced by mechanical sorting, collision, manual contact and other links. However, the existing technology generally uses ultrafine dry mist with a fixed particle size (such as 2-5μm) for spraying, which lacks the ability to adaptively control the surface state of fruits and vegetables. Especially when fruits and vegetables have serious fine cracks due to factors such as moisture fluctuations and mechanical extrusion, ultrafine dry mist is difficult to settle and penetrate crack gaps due to its small particle size, poor inertia and large disturbance, resulting in insufficient adhesion in the crack area, incomplete liquid film formation, and a significant decrease in the sterilization effect. There are problems such as uneven preservation coverage, weak protection of micro-wound wounds, and easy cross infection, which seriously restricts the stability and adaptability of this preservation method in complex dynamic environments.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0005] The purpose of the present invention is to provide a fruit and vegetable preservation method based on slightly acidic electrolyzed water dry mist, which introduces multispectral image acquisition and intelligent image recognition technology, combines a convolutional neural network model to accurately identify and quantitatively score the severity of microcracks on the surface of fruits and vegetables, and adjusts the particle size of the slightly acidic electrolyzed water dry mist in real time based on the scoring results, thereby realizing differentiated and dynamic preservation treatment of fruits and vegetables in different crack states, and can significantly improve the sedimentation and adhesion ability and liquid film formation efficiency of the dry mist in the crack area, enhance the local sterilization effect, and effectively make up for the technical defects of uneven coverage and insufficient preservation of traditional fixed particle size dry mist in the micro-wound wound area, and significantly improve the preservation accuracy and anti-pollution ability of fruits and vegetables under high-speed processing lines, so as to solve the problems in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solution: a method for preserving fruits and vegetables based on slightly acidic electrolyzed water dry mist, comprising the following steps:

[0007] The slightly acidic electrolyzed water is atomized into dry mist according to the set starting particle size range through the atomization control unit, and acts on the surface of fruits and vegetables during the high-speed assembly line transmission process to form a wide-area sterilization protection mist layer, thereby achieving preliminary sterilization and preservation of the entire surface of fruits and vegetables;

[0008] The multispectral image acquisition device is used to collect multispectral image information of the outer surface of fruits and vegetables in a non-contact manner while the fruits and vegetables pass through the production line;

[0009] The raw image data obtained from the multispectral imaging acquisition device usually contains interference factors such as noise and lighting changes. In order to obtain stable recognition accuracy, the image data needs to be preprocessed by denoising, normalization, brightness correction, and background removal. At the same time, the image frame is matched with the position of fruits and vegetables on the assembly line according to the assembly line speed and imaging frequency to ensure that the preprocessing results accurately correspond to the actual spatial positioning of fruits and vegetables;

[0010] Based on the preprocessed image data, the microcrack area is identified and located. The crack morphological features reflecting the severity of microcracks on the surface of fruits and vegetables are extracted from the microcrack area. After the extracted features are deeply analyzed through feature engineering technology, the analyzed features are used as multi-dimensional feature vectors to characterize the severity of microcracks on the surface of fruits and vegetables.

[0011] The feature vector is input into a convolutional neural network model that was previously trained offline and deployed in the field to score the severity of the crack based on its morphological characteristics.

[0012] According to the severity of the cracks output by the convolutional neural network, the working parameters of the atomizing nozzle are adjusted in real time to dynamically change the size of the dry mist particles. If severe cracks are detected on the surface of fruits and vegetables, the droplet size is automatically increased to enhance the sedimentation ability and adhesion of the droplets, ensuring that they can effectively enter the crack gaps and adhere to the inside of the cracks to form a liquid film, thereby enhancing the local sterilization and preservation effects. If the cracks are lighter or non-existent, a smaller particle size (such as 2-5μm) is maintained to increase the coverage area and reduce water consumption.

[0013] Preferably, collecting multispectral image information of the outer surface of fruits and vegetables by a multispectral image acquisition device specifically includes the following steps:

[0014] First, install the multispectral image acquisition device at the appropriate position of the fruit and vegetable processing line to ensure that the viewing angle covers the area where the fruits and vegetables pass through;

[0015] Secondly, when fruits and vegetables pass through the collection area, the trigger sensor synchronously starts the multi-spectral imaging device to continuously scan the surface of the fruits and vegetables;

[0016] Next, a multi-channel light source is used to illuminate the surface of fruits and vegetables frame by frame according to preset bands (such as visible light, near infrared, short-wave infrared, etc.), and an imaging device simultaneously collects reflection images of each band;

[0017] Subsequently, the collected multispectral image data is sent to the data processing unit to provide high-quality, multi-dimensional visual information support for subsequent analysis steps such as image preprocessing, feature extraction and crack identification.

[0018] Preferably, matching the image frame with the position of the fruit and vegetable on the assembly line according to the assembly line speed and the imaging frequency generally includes the following steps:

[0019] First, high-precision encoders or position sensors are installed at key locations on the assembly line to detect the conveyor belt speed and the time nodes when fruits and vegetables pass through in real time.

[0020] Secondly, the sampling frequency of the imaging system is synchronized with the speed data of the sensor so that each frame of the image corresponds to the position of the fruit and vegetable at a specific point in time;

[0021] Next, the theoretical spatial position of the fruits and vegetables on the conveyor belt corresponding to each frame of the image is calculated based on the linear speed of the conveyor belt and the image acquisition time interval;

[0022] Then, the actual position is fine-tuned and calibrated based on the trigger signal of fruit and vegetable entry or the contour information of fruit and vegetable detected in the image;

[0023] Finally, the image frame number is bound to the fruit and vegetable identification information to establish a one-to-one correspondence, ensuring that each image can accurately correspond to the actual fruit and vegetable individual in the preprocessing, feature extraction and subsequent control links, realizing the simultaneous alignment of image processing and physical position.

[0024] Preferably, the identification and location of microcrack areas based on the preprocessed image data can be carried out by combining multispectral image enhancement with morphological analysis. First, the contrast between the crack area and the normal fruit and vegetable skin in the spectral response is enhanced by band fusion or band difference, so that the microcracks are more prominent in a specific band combination; then, an edge detection algorithm (such as Canny or Laplacian operator) is used to extract the suspected crack edge, and then the image segmentation technology (such as regional growth, threshold segmentation or superpixel-based clustering) is combined to locate the preliminary crack area; then, the crack shape is optimized through morphological operations (such as corrosion, expansion, and refinement) to remove artifacts and redundant noise; finally, the target area that truly meets the microcrack characteristics is further screened out by combining geometric features and spatial positions to achieve high-precision identification and positioning.

[0025] Preferably, for the microcrack area, crack morphological features reflecting the severity of microcracks on the surface of fruits and vegetables are extracted therefrom, wherein the extracted crack morphological features include the degree of discreteness of cracks in the space on the surface of fruits and vegetables and the change in reflectivity of the crack area in different spectral bands. After in-depth analysis of the degree of discreteness of cracks in the space on the surface of fruits and vegetables and the change in reflectivity of the crack area in different spectral bands is performed through feature engineering technology, a crack distribution discrete index and a crack depth index are generated respectively, and the crack distribution discrete index and the crack depth index are used as multi-dimensional feature vectors to characterize the severity of microcracks on the surface of fruits and vegetables.

[0026] Preferably, a feature vector constructed by the crack distribution discrete index and the crack depth index is input into a convolutional neural network model that has been previously trained offline and deployed on site, and a crack structural abnormality is generated based on the convolutional neural network model, and the severity of the crack is scored by the crack structural abnormality.

[0027] Preferably, the specific steps of deeply analyzing the degree of crack dispersion on the surface space of fruits and vegetables by feature engineering technology to generate crack distribution dispersion index are as follows:

[0028] First, for all crack areas, mark the coordinates of the center point of each crack and construct a set of crack spatial positions. , where N represents the number of cracks detected in the image, is the geometric center coordinate of the i-th crack area. On this basis, the crack distance map is constructed ,in, , represents the Euclidean distance between the geometric centers of crack i and crack j, which is used to measure the spatial discreteness between cracks;

[0029] According to the crack spatial position set and crack distance diagram, the maximum connection dissipation ratio is introduced to calculate the crack distribution discrete index. The calculation expression is:

[0030]

[0031] in, is the crack distribution dispersion index, is a very small positive number to prevent the denominator from being zero; is the farthest distance between crack i and other cracks, indicating its most extreme discrete trend; It is the reciprocal sum of the distances between crack i and other cracks, and measures the overall tightness of their connection. The crack distribution discrete index combines the dual characteristics of local extreme value diffusion and global sparsity. The larger the value, the more dispersed the crack distribution, the more extensive the damage to the surface of fruits and vegetables, and the higher the severity of the cracks. It is suitable for severe fruits and vegetables with complex microcracks and widespread distribution.

[0032] Damage identification scenario.

[0033] Preferably, the specific steps of performing in-depth analysis on the reflectivity changes of the crack area in different spectral bands by feature engineering technology to generate the crack depth index are as follows:

[0034] First, in the preprocessed multispectral image, the pixel-level reflectance values ​​of the crack area in multiple specific bands are extracted and recorded as ,in represents the pixel position, is the kth band. Next, the local reflectivity gradient vector of the crack area in the band dimension is defined as:

[0035]

[0036] in, is the pixel position The local reflectivity gradient vector at reflects the reflectivity change trend of this point in multiple spectral bands. This vector describes the reflectivity change rate of the crack area between different bands (similar to the slope characteristics of the spectral curve), and n represents the total number of bands;

[0037] After obtaining the local reflectivity gradient vector of the crack area, the crack penetration index is calculated based on the tensor projection intensity of the local reflectivity gradient vector in the main extension direction of the crack. It is used to measure the consistency between the local spectral change trend and the main direction of the crack structure, and to characterize the penetration depth and development severity of the crack in the organizational structure. Assume that the main extension direction vector of the crack is , then the calculation expression of the crack penetration index in the crack area is:

[0038]

[0039] in: It is an indicator of crack penetration depth. represents the spatial extent of the crack region, is the projection value of the local reflectivity gradient vector on the main direction of the crack, To enhance the power factor of crack penetration variation, is the crack edge sharpness factor (such as the weight map of the second-order derivative enhancement of the edge), which is used to emphasize the area with more severe rupture.

[0040] Preferably, according to the severity of the crack output by the convolutional neural network, the working parameters of the atomizing nozzle are adjusted in real time to dynamically change the size of the dry mist particle size. The specific steps are as follows:

[0041] After obtaining the crack structure abnormality, first determine whether the difference between it and the reference threshold of the crack structure abnormality is positive, that is, compare whether the severity of the current cracks on the surface of fruits and vegetables exceeds the preset benchmark. For this purpose, define a difference function, and the expression of the difference function is:

[0042]

[0043] in, is the abnormality of the crack structure output by the convolutional neural network. The larger the value, the more serious the crack. is the reference threshold of crack structure abnormality, which is used to measure whether the crack reaches or exceeds the acceptable limit. is the difference between the crack structure abnormality and the reference threshold of the crack structure abnormality, which is used as a temporary judgment to determine whether the particle size needs to be increased. , indicating that the current crack condition is still within the safe range; if , it means that the abnormal degree of crack exceeds the benchmark, and the droplet size needs to be dynamically increased in the subsequent steps;

[0044] like , then keep the default particle size within the initial particle size range of the atomizing nozzle;

[0045] like , it is necessary to expand the initial particle size range to increase the droplet size appropriately and penetrate the cracks. The expression for expanding the initial particle size range is:

[0046]

[0047] in is the target droplet size, and The minimum and maximum allowable droplet sizes are designed for the system to limit the extreme values ​​of the final particle size, ensure the safety of the nozzle hardware and the stability of the spray effect. is the initial particle size range (standard atomization parameter used when cracks are not severe or do not exist), and its value is usually in the range of 2-5μm. Adjust the sensitivity coefficient for droplet size, used to enlarge or reduce The adjustment range brought about;

[0048] Achieve target droplet size After calculation, the system sends the target droplet size to the spray control unit, and updates the parameters such as the working voltage, frequency or nozzle geometry channel of the atomizing nozzle in real time, so that the actual sprayed dry fog particle size is adjusted to the target droplet size. When a high degree of abnormal cracks is detected, the nozzle will automatically adjust to the large particle size range to improve the droplet settling ability and adhesion, so that the droplets can penetrate into the crack gaps and form a relatively thick liquid film, enhancing sterilization and preservation; if the degree of cracks is mild, the default small particle size is maintained to achieve a balance between wide-area coverage and water saving. Through this closed-loop mechanism, the spray system can respond quickly on the assembly line and implement targeted preservation strategies, improve the protection effect of fruits and vegetables in micro-crack scenarios, and reduce cross-contamination and corruption risks.

[0049] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0050] The present invention introduces multispectral image acquisition and intelligent image recognition technology, combines with a convolutional neural network model to accurately identify and quantitatively score the severity of microcracks on the surface of fruits and vegetables, and adjusts the particle size of slightly acidic electrolyzed water dry mist in real time based on the scoring results, thereby realizing differentiated and dynamic preservation treatment of fruits and vegetables in different crack states. It can significantly improve the sedimentation and adhesion ability and liquid film formation efficiency of dry mist in the crack area, enhance the local bactericidal effect, and effectively make up for the technical defects of uneven coverage and insufficient preservation of traditional fixed-particle dry mist in the micro-wound area, and significantly improve the preservation accuracy and anti-pollution ability of fruits and vegetables under high-speed processing lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0052] Figure 1 The present invention is a method flow chart of the method for preserving fruits and vegetables based on slightly acidic electrolyzed water dry mist. DETAILED DESCRIPTION

[0053] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0054] The present invention provides Figure 1 The method for preserving fruits and vegetables based on slightly acidic electrolyzed water dry mist shown in the figure comprises the following steps:

[0055] The slightly acidic electrolyzed water is atomized into dry mist according to the set starting particle size range (e.g. 2-5 μm) through the atomization control unit, and acts on the surface of fruits and vegetables during the high-speed assembly line transmission process to form a wide-area sterilization protection mist layer, thereby achieving preliminary sterilization and preservation of the entire surface of fruits and vegetables;

[0056] The particle size setting is intended to achieve rapid droplet diffusion, high coverage and gasification efficiency in the initial stage of treatment, and serve as the initial reference standard for subsequent adaptive particle size adjustment, providing the system with basic freshness protection and continuous processing capabilities.

[0057] The multispectral image acquisition device is used to collect multispectral image information of the outer surface of fruits and vegetables in a non-contact manner while the fruits and vegetables pass through the production line;

[0058] The image acquisition system covers multiple bands including visible light, near infrared, and short-wave infrared. It can comprehensively capture information such as color texture, spectral reflectance changes, and microstructure disturbances on the surface of fruits and vegetables. In particular, it can identify microcrack areas that are invisible to the naked eye due to mechanical impact, water expansion, or epidermal tissue rupture, providing high-resolution original image data input for subsequent data analysis and judgment.

[0059] The multispectral image information of the outer surface of fruits and vegetables is collected by a multispectral image acquisition device, which specifically includes the following steps:

[0060] First, install the multispectral image acquisition device at the appropriate position of the fruit and vegetable processing line to ensure that the viewing angle covers the area where the fruits and vegetables pass through;

[0061] Secondly, when fruits and vegetables pass through the collection area, the trigger sensor synchronously starts the multi-spectral imaging device to continuously scan the surface of the fruits and vegetables;

[0062] Next, a multi-channel light source is used to illuminate the surface of fruits and vegetables frame by frame according to preset bands (such as visible light, near infrared, short-wave infrared, etc.), and an imaging device simultaneously collects reflection images of each band;

[0063] Subsequently, the collected multispectral image data is sent to the data processing unit to provide high-quality, multi-dimensional visual information support for subsequent analysis steps such as image preprocessing, feature extraction and crack identification.

[0064] The raw image data obtained from the multispectral imaging acquisition device usually contains interference factors such as noise and lighting changes. In order to obtain stable recognition accuracy, the image data needs to be preprocessed by denoising, normalization, brightness correction, and background removal. At the same time, the image frame is matched with the position of fruits and vegetables on the assembly line according to the assembly line speed and imaging frequency to ensure that the preprocessing results accurately correspond to the actual spatial positioning of fruits and vegetables;

[0065] Matching the image frame with the position of the fruit and vegetable on the assembly line according to the assembly line speed and imaging frequency usually includes the following steps:

[0066] First, high-precision encoders or position sensors are installed at key locations on the assembly line to detect the conveyor belt speed and the time nodes when fruits and vegetables pass through in real time.

[0067] Secondly, the sampling frequency of the imaging system is synchronized with the speed data of the sensor so that each frame of the image corresponds to the position of the fruit and vegetable at a specific point in time;

[0068] Next, the theoretical spatial position of the fruits and vegetables on the conveyor belt corresponding to each frame of the image is calculated based on the linear speed of the conveyor belt and the image acquisition time interval;

[0069] Then, the actual position is fine-tuned and calibrated based on the trigger signal of fruit and vegetable entry or the contour information of fruit and vegetable detected in the image;

[0070] Finally, the image frame number is bound to the fruit and vegetable identification information to establish a one-to-one correspondence, ensuring that each image can accurately correspond to the actual fruit and vegetable individual in the preprocessing, feature extraction and subsequent control links, realizing the simultaneous alignment of image processing and physical position.

[0071] Based on the preprocessed image data, the microcrack area is identified and located. The crack morphological features reflecting the severity of microcracks on the surface of fruits and vegetables are extracted from the microcrack area. After the extracted features are deeply analyzed through feature engineering technology, the analyzed features are used as multi-dimensional feature vectors to characterize the severity of microcracks on the surface of fruits and vegetables.

[0072] Based on the preprocessed image data, the identification and location of microcrack areas can be carried out by combining multispectral image enhancement with morphological analysis. First, the contrast between the crack area and the normal fruit and vegetable skin in the spectral response is enhanced by band fusion or band difference, so that the microcracks are more prominent in the specific band combination; then, the edge detection algorithm (such as Canny or Laplacian operator) is used to extract the suspected crack edge, and then the image segmentation technology (such as regional growth, threshold segmentation or superpixel-based clustering) is combined to locate the preliminary crack area; then, the crack shape is optimized through morphological operations (such as corrosion, expansion, and refinement) to remove artifacts and redundant noise points; finally, the target area that truly meets the microcrack characteristics is further screened out by combining geometric features and spatial positions to achieve high-precision identification and positioning.

[0073] For the microcrack area, crack morphological features reflecting the severity of microcracks on the surface of fruits and vegetables are extracted. The extracted crack morphological features include the discrete degree of cracks in the space on the surface of fruits and vegetables and the reflectivity changes of different spectral bands to the crack area. After in-depth analysis of the discrete degree of cracks in the space on the surface of fruits and vegetables and the reflectivity changes of different spectral bands to the crack area through feature engineering technology, the crack distribution discrete index and crack depth index are generated respectively. The crack distribution discrete index and crack depth index are used as multi-dimensional feature vectors to characterize the severity of microcracks on the surface of fruits and vegetables.

[0074] The high degree of spatial dispersion of cracks on the surface of fruits and vegetables usually indicates the presence of multiple independent and discontinuous microcrack areas on the surface of fruits and vegetables. This feature reflects that the overall epidermal structure of fruits and vegetables is damaged over a wide range, local stress is uneven, or the tissue is more fragile. Compared with concentrated cracks, discrete distribution often means that fruits and vegetables are affected by stress or mechanical disturbance at multiple points, and micro-damage appears in a multi-source and multi-point manner, which is more difficult to cover and repair through local preservation intervention, and also increases the risk of multi-point microbial invasion and cross-contamination. Therefore, from the perspective of preservation and quality stability, a highly discrete spatial distribution of cracks usually means that the cracks are more severe and the overall fragility of fruits and vegetables is more fragile, which should be given special attention and differentiated treatment.

[0075] The specific steps of deeply analyzing the discrete degree of cracks on the surface space of fruits and vegetables through feature engineering technology to generate crack distribution discrete index are as follows:

[0076] First, for all crack areas, mark the coordinates of the center point of each crack and construct a set of crack spatial positions. , where N represents the number of cracks detected in the image, is the geometric center coordinate of the i-th crack area. On this basis, the crack distance map is constructed ,in, , represents the Euclidean distance between the geometric centers of crack i and crack j, which is used to measure the spatial discreteness between cracks;

[0077] The role of this step is to structure the discrete distribution relationship in space into a quantitatively calculable distance network, laying the foundation for the construction of discrete indicators of crack distribution.

[0078] According to the crack spatial position set and crack distance diagram, the maximum connection dissipation ratio is introduced to calculate the crack distribution discrete index. The calculation expression is:

[0079]

[0080] in, is the crack distribution dispersion index, is a very small positive number to prevent the denominator from being zero; is the farthest distance between crack i and other cracks, indicating its most extreme discrete trend; It is the reciprocal sum of the distances between crack i and other cracks, measuring the overall tightness of their connection. The crack distribution discrete index combines the dual characteristics of local extreme value diffusion and global sparsity. The larger the index, the more dispersed the crack distribution, the more extensive the damage to the fruit and vegetable surface, and the higher the severity of the crack. It is suitable for severe damage identification scenarios with complex and widely distributed microcracks.

[0081] It can be seen from the crack distribution dispersion index that the larger the performance value of the crack distribution dispersion index generated by the feature engineering technology after in-depth analysis of the dispersion degree of cracks on the surface space of fruits and vegetables, the more serious the microcracks on the surface of fruits and vegetables. The crack distribution dispersion index comprehensively analyzes the distribution characteristics of cracks on the surface space of fruits and vegetables through feature engineering technology, and measures the diffusivity and dispersion between crack areas. When the cracks are widely distributed, discrete in position, and large in distance from each other, the crack distribution dispersion index will increase significantly, reflecting that the fruit and vegetable skin is stressed at multiple points, the structural integrity is reduced, and the microcracks are more serious; when the cracks are relatively concentrated or the number is sparse and the space is compact, the crack distribution dispersion index is low, indicating that the damage range of fruits and vegetables is limited and the overall skin is kept in good condition. Therefore, the level of the crack distribution dispersion index can effectively characterize the severity of the cracks and provide a reliable quantitative basis for the intelligent judgment of the fruit and vegetable preservation strategy.

[0082] The more significant the change in reflectivity of the crack area in different spectral bands, the more serious the microcracks on the surface of fruits and vegetables are. The reason is that the generation of microcracks will cause local changes in the surface structure, material and moisture distribution of fruits and vegetables, especially the phenomenon of depression, tissue damage, water seepage or cell fluid leakage often occurs inside the crack area. These changes will significantly affect the light reflection characteristics of the area in multiple bands such as visible light, near infrared or short-wave infrared. For example, water has a strong absorption capacity in the near-infrared band. If there is water seepage in the crack area, its reflectivity in the near-infrared band will decrease significantly; and structural damage causes changes in surface roughness, which will also produce a strong scattering effect in the visible band. Therefore, when multiple spectral channels show obvious differences in reflectivity changes in the crack area, it often means that the crack is not only a surface scratch, but is more likely to involve deep tissue damage below the skin of fruits and vegetables, reflecting a higher degree of crack severity.

[0083] The specific steps of using feature engineering technology to deeply analyze the reflectivity changes of the crack area in different spectral bands to generate crack penetration indicators are as follows:

[0084] First, in the preprocessed multispectral image, the pixel-level reflectance values ​​of the crack area in multiple specific bands are extracted and recorded as ,in represents the pixel position, is the kth band. Next, the local reflectivity gradient vector of the crack area in the band dimension is defined as:

[0085]

[0086] in, is the pixel position The local reflectivity gradient vector at reflects the reflectivity change trend of this point in multiple spectral bands. This vector describes the reflectivity change rate of the crack area between different bands (similar to the slope characteristics of the spectral curve), and n represents the total number of bands;

[0087] The purpose of this step is to establish a directional spectral variation feature to capture the cross-band reflectivity mutation behavior caused by structural fractures, water penetration, internal tissue exposure, etc., providing a basis for subsequent depth characterization.

[0088] After obtaining the local reflectivity gradient vector of the crack area, the crack penetration index is calculated based on the tensor projection intensity of the local reflectivity gradient vector in the main extension direction of the crack. It is used to measure the consistency between the local spectral change trend and the main direction of the crack structure, and to characterize the penetration depth and development severity of the crack in the organizational structure. Assume that the main extension direction vector of the crack is , then the calculation expression of the crack penetration index in the crack area is:

[0089]

[0090] in: It is an indicator of crack penetration depth. represents the spatial extent of the crack region, is the projection value of the local reflectivity gradient vector on the main direction of the crack, To enhance the power factor of crack penetration variation, is the crack edge sharpness factor (e.g., the weight map of the second-order derivative enhancement of the edge), which is used to emphasize the area with more severe rupture;

[0091] The more dramatic the spectral response changes along the crack direction, the more likely it is that there is deep structural damage, which can be used to accurately quantify the depth and severity of microcracks.

[0092] It can be seen from the crack penetration index that the larger the performance value of the crack penetration index generated by the feature engineering technology through in-depth analysis of the reflectivity changes of the crack area in different spectral bands, the more serious the microcracks on the surface of fruits and vegetables are. The crack penetration index is constructed based on the reflectivity change trend of the crack area under different spectral bands. It comprehensively considers the intensity, directionality and consistency of the spectral response with the main axis direction of the crack. When the crack is deeper, more complex, or has penetrated the peel to the internal tissue, it will cause the area to show more drastic and clear reflectivity changes under the multi-band spectrum, and then form a higher consistency tensor projection between the spectral gradient vector and the main direction of the crack, so that the crack penetration index increases. On the contrary, if the crack is shallow or a surface scratch, its effect on the spectral response is weak and has no obvious directionality, then the projection intensity is low, and the performance value of the crack penetration index is also reduced. Therefore, the crack penetration index can effectively quantify the structural severity of microcracks and is an important reference for intelligent assessment of crack risks.

[0093] The feature vector is input into a convolutional neural network model that was previously trained offline and deployed in the field to score the severity of the crack based on its morphological characteristics.

[0094] The feature vector constructed by the crack distribution discrete index and the crack depth index is input into the convolutional neural network model that has been trained offline in advance and deployed on site. The crack structure abnormality is generated based on the convolutional neural network model, and the severity of the crack is scored according to the crack structure abnormality.

[0095] The convolutional neural network model that is pre-trained offline and deployed on-site refers to the process of fine-tuning the convolutional neural network in a laboratory or computing platform using supervised learning in advance according to the specific application needs of surface crack identification of fruits and vegetables. Specifically, first, in the offline stage, a large amount of multispectral image sample data of the surface of fruits and vegetables is collected, and these image data are preprocessed and manually annotated, including marking the location, type and severity of the cracks, to form a data set with high-quality annotation information; then, multiple morphological features of the cracks (such as crack distribution discrete index, crack depth index, etc.) are used as input features to construct feature vectors, and they are associated with the corresponding annotation information (such as crack severity level or specific value) to generate standard training sample pairs; next, a suitable convolutional neural network structure (such as LeNet, ResNet or Inception series network) is selected, and according to the crack feature data and corresponding annotations, a server with powerful GPU or CPU computing power is used to perform long-term, multi-round iterative training and parameter adjustment to achieve the optimal configuration of network structure, convolution kernel weights and parameters; when the model performance reaches the ideal recognition accuracy and generalization ability on an independent test set, the offline training process of the convolutional neural network model is completed, and efficient crack recognition network parameters suitable for field use are obtained.

[0096] Deployment on site refers to the process of transferring the above-mentioned offline trained convolutional neural network model to the fruit and vegetable processing line site, and actually running it with the help of embedded computing devices, edge computing devices or industrial-grade computing units (such as edge servers, industrial computers, etc.). In actual on-site applications, real-time multispectral image acquisition equipment is installed on the fruit and vegetable production line. When fruits and vegetables pass by, image acquisition and preprocessing operations can be completed quickly. The crack distribution discrete index and crack penetration index characteristic parameters are extracted from the image data collected on site in real time, and then the pre-deployed convolutional neural network model is used to process and analyze these feature vectors in real time to quickly obtain the crack structure abnormality. At this time, the convolutional neural network model no longer needs to be trained on a large scale again, but only needs to complete efficient forward propagation operations to quickly output the fruit and vegetable crack structure abnormality, and quickly and accurately evaluate the severity of the crack through the crack structure abnormality. This offline training and online deployment strategy not only ensures that the model has high recognition accuracy and generalization performance, but also effectively avoids the computing burden and response delay caused by on-site retraining, thereby greatly improving the real-time and stability of crack identification and intelligent decision-making in actual industrial applications, and realizing dynamic crack identification and real-time assessment of crack severity.

[0097] The convolutional neural network is not specifically limited here, and can realize the discrete index of crack distribution. and fracture depth index Comprehensive analysis to generate crack structure anomaly In order to realize the technical solution of the present invention, the present invention provides a specific implementation method; the abnormality of the crack structure The generated expression is:

[0098] , where , They are crack distribution dispersion index and fracture depth index The preset scaling factor of , The preset proportionality coefficient refers to the ratio of two indicators of different dimensions or dimensions (i.e., crack distribution discreteness index) when constructing the expression of crack structure abnormality. and fracture depth index ) to perform reasonable weighted fusion, which requires artificially pre-set weight factors, which are and The role of these two coefficients is to adjust and balance the weights of the crack distribution dispersion index and the crack depth index in the final abnormality score. and There may be differences in the numerical range, change trend and influencing mechanism. If they are directly added, it is easy to be distorted. Therefore, it is necessary to normalize or compensate for the difference through the preset proportional coefficient so that the contribution of the two in the output of the convolutional neural network conforms to the physical or statistical meaning of the actual crack severity. Usually these coefficients are obtained through experiments or optimization in the training stage, or set by empirical values, and , All are greater than 0, ensuring the interpretability and stability of the model.

[0099] It can be seen from the crack structure abnormality that the greater the performance value of the crack distribution discrete index generated after in-depth analysis of the discrete degree of cracks in the surface space of fruits and vegetables through feature engineering technology, the greater the performance value of the crack depth index generated by in-depth analysis of the reflectivity changes of the crack area in different spectral bands through feature engineering technology, that is, the greater the performance value of the crack structure abnormality generated when predicting the severity of cracks through the convolutional neural network model trained offline in advance and deployed on site, the more serious the microcracks on the surface of fruits and vegetables are, and vice versa.

[0100] According to the severity of the cracks output by the convolutional neural network, the working parameters of the atomizer nozzle are adjusted in real time to dynamically change the size of the dry mist particles. If severe cracks are detected on the surface of fruits and vegetables, the droplet size is automatically increased to enhance the settling ability and adhesion of the droplets, ensuring that they can effectively enter the crack gaps and adhere to the inside of the cracks to form a liquid film, thereby enhancing the local sterilization and preservation effects; if the cracks are light or non-existent, a smaller particle size (such as 2-5μm) is maintained to increase the coverage area and reduce water consumption;

[0101] Through the above-mentioned adaptive control mechanism, the directional enhanced preservation of the cracked areas on the surface of fruits and vegetables can be achieved in high-speed assembly line operations, which can significantly reduce the risks of micro-wound infection and cross-contamination, and extend the storage and transportation life of fruits and vegetables.

[0102] According to the crack severity output by the convolutional neural network, the working parameters of the atomizer nozzle are adjusted in real time to dynamically change the dry fog particle size. The specific steps are as follows:

[0103] After obtaining the crack structure abnormality, first determine whether the difference between it and the reference threshold of the crack structure abnormality is positive, that is, compare whether the severity of the current cracks on the surface of fruits and vegetables exceeds the preset benchmark. For this purpose, define a difference function, and the expression of the difference function is:

[0104]

[0105] in, is the abnormality of the crack structure output by the convolutional neural network. The larger the value, the more serious the crack. is the reference threshold of crack structure abnormality, which is used to measure whether the crack reaches or exceeds the acceptable limit. is the difference between the crack structure abnormality and the reference threshold of the crack structure abnormality, which is used as a temporary judgment to determine whether the particle size needs to be increased. , indicating that the current crack condition is still within the safe range; if , it means that the abnormal degree of crack exceeds the benchmark, and the droplet size needs to be dynamically increased in the subsequent steps;

[0106] The purpose of this step is to quickly identify cracks on the surface of fruits and vegetables, avoid unnecessary adjustment of droplet size, and effectively improve the system response efficiency.

[0107] like , then keep the default particle size within the initial particle size range of the atomizing nozzle;

[0108] like , it is necessary to expand the initial particle size range to increase the droplet size appropriately and penetrate the cracks. The expression for expanding the initial particle size range is:

[0109]

[0110] in, is the target droplet size, and The minimum and maximum allowable droplet sizes are designed for the system to limit the extreme values ​​of the final particle size, ensure the safety of the nozzle hardware and the stability of the spray effect. is the initial particle size range (standard atomization parameter used when cracks are not severe or do not exist), and its value is usually in the range of 2-5μm. Adjust the sensitivity coefficient for droplet size, used to enlarge or reduce The adjustment range brought about;

[0111] This step can be achieved When the droplet size is larger, the droplet size increases exponentially, but still within The particle size is limited within the range to ensure that the particle size is not over-enlarged or below the usable value.

[0112] Achieve target droplet size After calculation, the system sends the target droplet size to the spray control unit, and updates the parameters such as the working voltage, frequency or nozzle geometry channel of the atomizing nozzle in real time, so that the actual sprayed dry fog particle size is adjusted to the target droplet size. When a high degree of abnormal cracks is detected, the nozzle will automatically adjust to the large particle size range to improve the droplet settling ability and adhesion, so that the droplets can penetrate into the crack gaps and form a relatively thick liquid film, enhancing sterilization and preservation; if the degree of cracks is mild, the default small particle size is maintained to achieve a balance between wide-area coverage and water saving. Through this closed-loop mechanism, the spray system can respond quickly on the assembly line and implement targeted preservation strategies, improve the protection effect of fruits and vegetables in micro-crack scenarios, and reduce cross-contamination and corruption risks.

[0113] The present invention introduces multispectral image acquisition and intelligent image recognition technology, combines with a convolutional neural network model to accurately identify and quantitatively score the severity of microcracks on the surface of fruits and vegetables, and adjusts the particle size of slightly acidic electrolyzed water dry mist in real time based on the scoring results, thereby realizing differentiated and dynamic preservation treatment of fruits and vegetables in different crack states, which can significantly improve the sedimentation and adhesion ability and liquid film formation efficiency of dry mist in the crack area, enhance the local bactericidal effect, and effectively make up for the technical defects of uneven coverage and insufficient preservation of traditional fixed-particle dry mist in the micro-wound area, thereby significantly improving the preservation accuracy, anti-pollution ability and commodity stability of fruits and vegetables under high-speed processing lines, and has good intelligence, flexibility and application promotion value.

[0114] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0115] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0116] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for preserving fruits and vegetables based on slightly acidic electrolyzed water dry mist, characterized in that: The following steps are involved: Atomize slightly acidic electrolyzed water into dry mist according to the set initial particle size range, and act on the surface of fruits and vegetables during high-speed assembly line transmission to form a wide-area sterilization protection mist layer; The multispectral image acquisition device is used to collect multispectral image information of the outer surface of the fruits and vegetables while they are passing through the production line; Process the collected raw images and match the image frames with the positions of fruits and vegetables on the assembly line according to the assembly line speed and imaging frequency; Based on the preprocessed image data, the microcrack area is identified and located. The crack morphological features reflecting the severity of microcracks on the surface of fruits and vegetables are extracted from the microcrack area. After the extracted features are deeply analyzed through feature engineering technology, the analyzed features are used as multi-dimensional feature vectors to characterize the severity of microcracks on the surface of fruits and vegetables. The feature vector is input into a convolutional neural network model that was previously trained offline and deployed in the field to score the severity of the crack based on its morphological characteristics. According to the crack severity output by the convolutional neural network, the working parameters of the atomizer nozzle are adjusted in real time to dynamically change the size of the dry mist particles.

2. The method for preserving fruits and vegetables based on slightly acidic electrolyzed water dry mist according to claim 1, characterized in that: The multispectral image information of the outer surface of fruits and vegetables is collected by a multispectral image acquisition device, which specifically includes the following steps: Install multispectral image acquisition devices on the fruit and vegetable processing line to ensure that the viewing angle covers the area where the fruits and vegetables pass through; When fruits and vegetables pass through the collection area, the trigger sensor synchronously starts the multispectral imaging device to continuously scan the surface of the fruits and vegetables; Use a multi-channel light source to illuminate the surface of fruits and vegetables frame by frame according to the preset bands, and use an imaging device to synchronously collect reflection images of each band; The collected multispectral image data is sent to the data processing unit.

3. The method for preserving fruits and vegetables based on slightly acidic electrolyzed water dry mist according to claim 1, characterized in that: Matching the image frame with the position of the fruit and vegetable on the pipeline according to the pipeline speed and imaging frequency includes the following steps: Install high-precision encoders at key locations on the assembly line to detect the conveyor belt speed and the time when fruits and vegetables pass through in real time; Synchronize the sampling frequency of the imaging system with the speed data of the sensor so that each frame of the image corresponds to the position of the fruit and vegetable at a point in time; According to the linear speed of the conveyor belt and the image acquisition time interval, the theoretical spatial position of the fruits and vegetables corresponding to each frame of the image on the conveyor belt is calculated; Combined with the contour information of fruits and vegetables detected in the image, the actual position is fine-tuned and calibrated; Bind the image frame number with the fruit and vegetable identification information to establish a one-to-one correspondence.

4. The method for preserving fruits and vegetables based on slightly acidic electrolyzed water dry mist according to claim 1, characterized in that: The microcrack area is identified and located based on the preprocessed image data by combining multispectral image enhancement with morphological analysis. The specific steps are as follows: The contrast between the crack area and the normal fruit and vegetable skin in spectral response is enhanced by band fusion, making the microcracks more prominent in a specific band combination; The edge detection algorithm is used to extract the suspected crack edge, and then the image segmentation technology is combined to locate the preliminary crack area; Optimize the crack shape through morphological operations to remove artifacts and redundant noise; The target areas that meet the microcrack characteristics are further screened out by combining geometric features and spatial positions to achieve high-precision identification and positioning.

5. The method for preserving fruits and vegetables based on slightly acidic electrolyzed water dry mist according to claim 1, characterized in that: For the microcrack area, crack morphological features reflecting the severity of microcracks on the surface of fruits and vegetables are extracted. The extracted crack morphological features include the discrete degree of cracks in the space on the surface of fruits and vegetables and the reflectivity changes of different spectral bands to the crack area. After in-depth analysis of the discrete degree of cracks in the space on the surface of fruits and vegetables and the reflectivity changes of different spectral bands to the crack area through feature engineering technology, the crack distribution discrete index and crack depth index are generated respectively. The crack distribution discrete index and crack depth index are used as multi-dimensional feature vectors to characterize the severity of microcracks on the surface of fruits and vegetables.

6. The method for preserving fruits and vegetables based on slightly acidic electrolyzed water dry mist according to claim 5, characterized in that: The feature vector constructed by the crack distribution discrete index and the crack depth index is input into the convolutional neural network model that has been trained offline in advance and deployed on site. The crack structure abnormality is generated based on the convolutional neural network model, and the severity of the crack is scored according to the crack structure abnormality.

7. The method for preserving fruits and vegetables based on slightly acidic electrolyzed water dry mist according to claim 5, characterized in that: The specific steps of deeply analyzing the discrete degree of cracks on the surface space of fruits and vegetables through feature engineering technology to generate crack distribution discrete index are as follows: For all crack regions, mark the coordinates of the center point of each crack and construct a set of crack spatial positions; Construct a crack distance map to measure the spatial discreteness between cracks; According to the crack spatial position set and crack distance diagram, the maximum connection dissipation ratio is introduced to calculate the crack distribution discrete index.

8. The method for preserving fruits and vegetables based on slightly acidic electrolyzed water dry mist according to claim 5, characterized in that: The specific steps of using feature engineering technology to deeply analyze the reflectivity changes of the crack area in different spectral bands to generate crack penetration indicators are as follows: In the preprocessed multispectral image, the pixel-level reflectivity values ​​of the crack area in multiple specific bands are extracted, and the local reflectivity gradient vector of the crack area in the band dimension is defined; The crack penetration index is calculated based on the tensor projection intensity of the local reflectivity gradient vector in the main extension direction of the crack. It is used to measure the consistency between the local spectral change trend and the main direction of the crack structure, and to characterize the penetration depth and development severity of the crack in the organizational structure.

9. The method for preserving fruits and vegetables based on slightly acidic electrolyzed water dry mist according to claim 5, characterized in that: According to the crack severity output by the convolutional neural network, the working parameters of the atomizer nozzle are adjusted in real time to dynamically change the dry fog particle size. The specific steps are as follows: Define a difference function to determine whether the difference between the crack structure abnormality and the reference threshold of the crack structure abnormality is positive, that is, to compare whether the severity of the current cracks on the surface of fruits and vegetables exceeds the preset benchmark. The expression of the difference function is: ； in, is the abnormality of the crack structure output by the convolutional neural network, is the reference threshold of crack structure abnormality, is the difference between the crack structure abnormality and the reference threshold of the crack structure abnormality, which is used to determine whether the particle size needs to be increased; like , then keep the default particle size within the initial particle size range of the atomizing nozzle; like , then the initial particle size range is expanded to make the droplet size increase appropriately. The expression for expanding the initial particle size range is: ； in, is the target droplet size, and The minimum and maximum allowable droplet sizes are used to limit the extreme values ​​of the final particle size. is the initial particle size range, Adjust the sensitivity coefficient for droplet size, used to enlarge or reduce The adjustment range brought about; Achieve target droplet size After calculation, the target droplet size is sent to the spray control unit so that the actual sprayed dry fog particle size is adjusted to the target droplet size.

Images