A method and system for controlling the cold storage and preservation of dried fruit

Abstract

This invention discloses a method and system for controlling the cold storage and preservation of dried fruit, relating to the field of optimization control technology. The method includes: acquiring the surface humidity value of the dried fruit immediately after it is placed in a cold storage and the humidity range of the cold storage; comparing the surface humidity value of the dried fruit with the humidity range of the cold storage; performing real-time monitoring of the dried fruit in the cold storage to obtain dried fruit images; preprocessing the dried fruit images to obtain images of mold spots and sugar bloom; performing image enhancement on the mold spot images to obtain target dried fruit images; inputting the target dried fruit images into an anomaly recognition model to obtain recognition results; and marking dried fruit with mold spots in the recognition results and transmitting them to the backend for processing. This invention improves the accuracy of mold spot recognition by distinguishing between mold spots and sugar bloom, thereby improving the preservation quality of dried fruit.

Description

Technical Field

[0001] This invention belongs to the field of optimization control technology, specifically relating to a method and system for controlling the cold storage and preservation of dried fruit. Background Technology

[0002] With the rapid development of the dried fruit industry, refrigeration has become one of the main methods of preserving dried fruit. In the existing refrigeration system, energy-saving control is mainly achieved through refrigeration units and dehumidifiers. By adopting optimized control strategies, temperature and humidity can be optimized to achieve precise temperature control and energy saving in dried fruit refrigeration.

[0003] The existing technology (publication number: CN121474810A) discloses a dynamic energy-saving control method and system for cold storage based on deep learning and multi-source data fusion. It relates to the field of cold storage energy-saving control technology. The method constructs a thermal field perception network using a fixed sparse sensor array and a mobile thermal field scanning robot to collect temperature, humidity, and thermal imaging data. After registration and interpolation, a three-dimensional thermal pixel field is reconstructed and spatiotemporally aligned with equipment parameters and inventory thermal properties. A digital twin is constructed based on a physical information neural network to deduce the thermal field evolution, and a spatiotemporal graph attention network is used to identify high-temperature hotspots and low-temperature safe zones. With the goal of minimizing electricity costs and controlling over-temperature risks, a multi-agent deep deterministic strategy gradient algorithm is used to solve for the optimal control strategy, including vent parameters and compressor frequency, driving the equipment to execute targeted cooling. The system includes a thermal field perception layer, a data fusion layer, a digital twin and prediction layer, a decision optimization layer, and an execution layer, achieving precise temperature control and energy saving.

[0004] The aforementioned patent achieves precise control of the cold storage and reduces refrigeration energy consumption through an optimal control strategy; however, in actual dried fruit preservation control, mold spots or sugar frosting may appear on the surface of the dried fruit. The two are quite similar in color and shape, making it impossible to accurately determine whether the dried fruit is fresh. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that mold spots or sugar frost appear on the surface of dried fruit in actual dried fruit preservation control. The two are similar in color and shape, making it impossible to accurately determine whether the dried fruit is fresh. Therefore, this invention proposes a method and system for controlling the cold storage and preservation of dried fruit.

[0006] In a first aspect of this invention, a method for controlling the cold storage and preservation of dried fruit is first proposed, the method comprising: Obtain the surface humidity value of the dried fruit immediately after it is placed in the cold storage and the humidity value range of the cold storage; the humidity value range includes a lower limit and an upper limit. The humidity value of the dried fruit surface is compared with the humidity range of the cold storage. If the humidity value of the dried fruit surface is greater than the upper limit of the range, the dried fruit surface is dehumidified until it is within the humidity range. If the humidity value of the dried fruit surface is less than the upper limit of the range, the dried fruit surface is moisturized until it is within the humidity range. Then, real-time monitoring of the dried fruit placed in the cold storage is used to obtain images of the dried fruit; The dried fruit images are preprocessed to obtain dried fruit mold images and dried fruit sugar frosting images; Image enhancement is performed on images of mold spots on dried fruit to obtain the target image of dried fruit; The target dried fruit image is substituted into the anomaly recognition model to obtain the recognition result; the recognition result is the location and size of the mold spots on the dried fruit. The dried fruit with mold spots in the identification results are marked and transmitted to the backend for processing.

[0007] Optionally, according to claim 1, the method for controlling the cold storage and preservation of dried fruit is characterized in that the dried fruit image is preprocessed to obtain dried fruit mold image and dried fruit sugar frosting image, specifically including: Convert the dried fruit image from RGB format to HSV format; The pixel values ​​of the H and S channels in the converted dried image are extracted and compared with the preset mold color threshold range and sugar frost color threshold range, respectively. Pixel regions that meet the mold color threshold range are divided into initial mold regions, and pixel regions that meet the frosting color threshold range are divided into initial frosting regions. Morphological opening operations were performed on the initial moldy area and the initial sugar frosting area to obtain the dried fruit moldy image and the dried fruit sugar frosting image, respectively; The specific formulas for the morphological opening operation include: in, Represents the image after morphological opening operation. The binary image obtained from the initial segmentation is represented by a pixel value of 0 or 1, where 1 represents a candidate region and 0 represents a background region; E represents the structuring element, x and y represent pixel coordinates, r represents the radius of the structuring element in pixels, and is used to define the shape and size of the neighborhood in morphological operations; m and n represent the offset coordinates in the structuring element, where m is the offset in the row direction and n is the offset in the column direction. This represents the erosion operator. This represents the dilation operator, where i and j represent the pixel coordinates of the current image.

[0008] Optionally, image enhancement of the dried fruit mold image to obtain the target dried fruit image includes: The enhanced mold image is subjected to adaptive gamma correction to obtain a first enhanced mold image; the second enhanced mold image is an image with enhanced mold edge details. The target dried fruit image is obtained by calculating the first enhanced moldy image using the multi-scale Retinex algorithm.

[0009] Optionally, the anomaly detection model is obtained by modifying the YOLOv8 model, specifically including: In the main structure and neck structure, the connection between the SPPF module and the Upsample module is disconnected, and an enhanced feature extraction module is added after the SPPF module and connected to the Upsample module. In the detection head, the original three Detect modules are replaced with the Detect_1 module, which is then connected to the neck structure C2f module. The working principle of the enhanced feature extraction module is as follows: The input to the enhanced feature extraction module is determined as the input feature map; The input feature map is subjected to horizontal average pooling and vertical average pooling to obtain a horizontal feature map and a vertical feature map, respectively. The horizontal feature map is segmented to obtain multiple sub-horizontal feature maps; The vertical feature map is segmented to obtain multiple sub-vertical feature maps; The first feature map is obtained by convolving the multiple sub-horizontal feature maps and then concatenating them; the second feature map is obtained by convolving the multiple sub-vertical feature maps and then concatenating them. The first and second feature maps are normalized and then concatenated to obtain the third feature map. The third feature map is subjected to depthwise convolution to obtain the fourth feature map; The fourth feature is normalized to obtain the fifth feature map; the third feature map and the fifth feature map are multiplied element by element to obtain the target feature map.

[0010] Optionally, the working principle of the Detect_1 module specifically includes: The three output features of the neck structure are respectively determined as three input features; the three input features are the first feature, the second feature, and the third feature; The first feature is upsampled to obtain the fourth feature, and the fourth feature is convolved to obtain the first convolutional feature; The second feature and the first convolutional feature are fused to obtain the first fused feature; The second feature is upsampled to obtain the fifth feature, and the fifth feature is convolved to obtain the second convolutional feature; The third feature and the second convolutional feature are fused to obtain the second fused feature; The first fusion feature is upsampled and then concatenated with the second fusion feature to obtain the concatenated feature. The concatenated feature is then substituted into the Detect module to obtain the recognition result.

[0011] In a second aspect of this invention, a dried fruit cold storage and preservation control system is provided, the system comprising: Data acquisition module: acquires the surface humidity value of dried fruit immediately after it is placed in the cold storage and the humidity value range of the cold storage; the humidity value range includes a lower limit and an upper limit. Humidity control module: compares the surface humidity value of the dried fruit with the humidity range of the cold storage; if the surface humidity value of the dried fruit is greater than the upper limit of the range, the surface of the dried fruit is dehumidified until it is within the humidity range; if the surface humidity value of the dried fruit is less than the upper limit of the range, the surface of the dried fruit is moisturized until it is within the humidity range. Image acquisition module: Real-time monitoring of dried fruit placed in cold storage to obtain images of dried fruit; Preprocessing module: preprocesses the dried fruit image to obtain dried fruit mold image and dried fruit sugar frosting image; Image enhancement module: Enhances the image of mold spots on dried fruit to obtain the target dried fruit image; Mold Spot Recognition Module: Substitutes the target dried fruit image into the anomaly recognition model to obtain the recognition result; the recognition result is the location and size of the mold spots on the dried fruit. Backend processing module: Marks the dried fruit with mold spots in the recognition results and transmits them to the backend for processing.

[0012] Optionally, the preprocessing module includes: a conversion module, a tone comparison module, an initial frosting region module, and a morphological opening module. The conversion module is used to convert the dried fruit image from RGB format to HSV format; The tone comparison module is used to extract the pixel values ​​of the H channel and S channel in the dried image after conversion, and compare them with the preset mold tone threshold range and sugar frosting tone threshold range, respectively. The initial frosting region module is used to divide the pixel region that meets the mold spot hue threshold range into an initial mold spot region, and to divide the pixel region that meets the frosting hue threshold range into an initial frosting region. The morphological opening module is used to perform morphological opening operations on the initial moldy area and the initial sugar frosting area to obtain the dried fruit moldy image and the dried fruit sugar frosting image, respectively. The specific formulas for the morphological opening operation include: in, Represents the image after morphological opening operation. The binary image obtained from the initial segmentation is represented by a pixel value of 0 or 1, where 1 represents a candidate region and 0 represents a background region; E represents the structuring element, x and y represent pixel coordinates, r represents the radius of the structuring element in pixels, and is used to define the shape and size of the neighborhood in morphological operations; m and n represent the offset coordinates in the structuring element, where m is the offset in the row direction and n is the offset in the column direction. This represents the erosion operator. This represents the dilation operator, where i and j represent the pixel coordinates of the current image.

[0013] Optionally, the image enhancement module is also used for: a gamma correction module and a multi-scale Retinex algorithm module. The gamma correction module is used to apply adaptive gamma correction to the enhanced mold spot image to obtain a first enhanced mold spot image; the second enhanced mold spot image is an image that enhances the edge details of the mold spot. The multi-scale Retinex algorithm module is used to calculate the target dried fruit image from the first enhanced moldy image using the multi-scale Retinex algorithm.

[0014] Optionally, the mold spot recognition module is also used for: The anomaly detection model is a modification of the YOLOv8 model, specifically including: In the main structure and neck structure, the connection between the SPPF module and the Upsample module is disconnected, and an enhanced feature extraction module is added after the SPPF module and connected to the Upsample module. In the detection head, the original three Detect modules are replaced with the Detect_1 module, which is then connected to the neck structure C2f module. The working principle of the enhanced feature extraction module is as follows: The input to the enhanced feature extraction module is determined as the input feature map; The input feature map is subjected to horizontal average pooling and vertical average pooling to obtain a horizontal feature map and a vertical feature map, respectively. The horizontal feature map is segmented to obtain multiple sub-horizontal feature maps; The vertical feature map is segmented to obtain multiple sub-vertical feature maps; The first feature map is obtained by convolving the multiple sub-horizontal feature maps and then concatenating them; the second feature map is obtained by convolving the multiple sub-vertical feature maps and then concatenating them. The first and second feature maps are normalized and then concatenated to obtain the third feature map. The third feature map is subjected to depthwise convolution to obtain the fourth feature map; The fourth feature is normalized to obtain the fifth feature map; the third feature map and the fifth feature map are multiplied element by element to obtain the target feature map.

[0015] Optionally, the working principle of the Detect_1 module specifically includes: The three output features of the neck structure are respectively determined as three input features; the three input features are the first feature, the second feature, and the third feature; The first feature is upsampled to obtain the fourth feature, and the fourth feature is convolved to obtain the first convolutional feature; The second feature and the first convolutional feature are fused to obtain the first fused feature; The second feature is upsampled to obtain the fifth feature, and the fifth feature is convolved to obtain the second convolutional feature; The third feature and the second convolutional feature are fused to obtain the second fused feature; The first fusion feature is upsampled and then concatenated with the second fusion feature to obtain the concatenated feature. The concatenated feature is then substituted into the Detect module to obtain the recognition result.

[0016] The beneficial effects of this invention are: This invention proposes a method and system for controlling the cold storage and preservation of dried fruit. By precisely controlling the humidity of the cold storage to the suitable storage range for dried fruit, condensation on the surface of the dried fruit can be effectively reduced, the risk of mold growth can be lowered, and the preservation quality can be improved. At the same time, by differentiating between mold spots and sugar frosting images, enhancing the mold spot images, and combining them with an anomaly recognition model, the accuracy of mold spot recognition can be significantly improved. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 A flowchart of a method for controlling the cold storage and preservation of dried fruit provided in an embodiment of the present invention; Figure 2 This is a network structure diagram of an anomaly recognition model provided in an embodiment of the present invention; Figure 3 A network structure diagram of a YOLOv8 model provided in an embodiment of the present invention; Figure 4 This is a framework diagram of a dried fruit cold storage and preservation control system provided in an embodiment of the present invention. Detailed Implementation

[0019] 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.

[0020] 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.

[0021] This invention provides a method for controlling the cold storage and preservation of dried fruit. See also... Figure 1 , Figure 1 A flowchart illustrating a method for controlling the cold storage and preservation of dried fruit according to an embodiment of the present invention. The method includes the following steps: Obtain the surface humidity value of the dried fruit immediately after it is placed in the cold storage and the humidity range of the cold storage; the humidity range includes the lower limit and the upper limit of the range. The humidity value of the dried fruit surface is compared with the humidity range of the cold storage. If the humidity value of the dried fruit surface is greater than the upper limit of the range, the dried fruit surface is dehumidified until it is within the humidity range. If the humidity value of the dried fruit surface is less than the upper limit of the range, the dried fruit surface is moisturized until it is within the humidity range. Then, real-time monitoring of the dried fruit placed in the cold storage is used to obtain images of the dried fruit; Preprocessing the dried fruit images yields images of dried fruit mold spots and dried fruit frosting. Image enhancement is performed on images of mold spots on dried fruit to obtain the target image of dried fruit; The target dried fruit image is fed into the anomaly detection model to obtain the detection result; the detection result is the location and size of the mold spots on the dried fruit. The dried fruit with mold spots in the identification results are marked and transmitted to the backend for processing.

[0022] The present invention provides a method for controlling the cold storage and preservation of dried fruit. By controlling the humidity of dried fruit within the humidity range of the cold storage, the condensation of water droplets on dried fruit is reduced, the occurrence of mold spots is decreased, and the preservation quality of dried fruit is improved. By comparing and differentiating images of mold spots and images of sugar frosting on dried fruit, and by performing image enhancement on images of mold spots, the accuracy of mold spot identification is improved. An anomaly identification model is used to improve the accuracy of mold spot identification.

[0023] Specifically, the humidity range of the cold storage is set based on the staff's historical experience; the dried fruit with mold spots in the identification results is marked and transmitted to the backend for processing; specifically, the location and size of the mold spots on the dried fruit are obtained through the identification results and sent to the backend for processing. The specific processing may be: taking the location of the moldy dried fruit as the center, setting a preset radius area as the stale area and processing it.

[0024] In one implementation, the dried fruit image is preprocessed to obtain a dried fruit mold image and a dried fruit frosting image, specifically including: Convert the dried fruit image from RGB format to HSV format; extract the pixel values ​​of the H channel and S channel in the converted dried fruit image, and compare them with the preset mold color threshold range and sugar frost color threshold range, respectively. Pixel regions that meet the mold color threshold range are divided into initial mold regions, and pixel regions that meet the frosting color threshold range are divided into initial frosting regions. Morphological opening operations were performed on the initial moldy area and the initial sugar frosting area to obtain the dried fruit moldy area image and the dried fruit sugar frosting image, respectively. The specific formulas for morphological opening operations include: in, Represents the image after morphological opening operation. The binary image obtained from the initial segmentation is represented by a pixel value of 0 or 1, where 1 represents a candidate region and 0 represents a background region; E represents the structuring element, x and y represent pixel coordinates, r represents the radius of the structuring element in pixels, and is used to define the shape and size of the neighborhood in morphological operations; m and n represent the offset coordinates in the structuring element, where m is the offset in the row direction and n is the offset in the column direction. This represents the erosion operator. This represents the dilation operator, where i and j represent the pixel coordinates of the current image.

[0025] In one implementation, the dried fruit image is first converted from RGB format to HSV format. RGB format is easily affected by light intensity, which is not conducive to color-based segmentation of mold spots and sugar frosting. The hue threshold range for the mold spot region is pre-set to [Hm_min, Hm_max], typically taking values ​​from

[20] . ∘ 50 ∘ The hue threshold range for the frosting region is [Hsugar_min, Hsugar_max], typically set to [0, 0, 0]. ∘ 15 ∘ ]∪[345 ∘ 360 ∘ The image is divided into white and light yellow regions. The H and S channels of the HSV image are extracted, and each pixel is judged as follows: if the H value of a pixel is within the mold hue threshold range and the S value is greater than the preset sugar frost saturation threshold (e.g., S>0.2), then the pixel is marked as the initial mold hue region; if the H value of a pixel is within the sugar frost hue threshold range and the S value is less than the preset sugar frost saturation threshold (e.g., S<0.15), then the pixel is marked as the initial sugar frost region. After morphological opening operation, the problem of isolated noise points caused by the surface texture of dried fruit or light reflection is solved.

[0026] Optionally, image enhancement of the dried fruit mold image to obtain the target dried fruit image includes: Adaptive gamma correction is applied to the enhanced mold patch image to obtain the first enhanced mold patch image; the second enhanced mold patch image is an image with enhanced edge details of the mold patch. The target dried fruit image is obtained by calculating the first enhanced moldy image using the multi-scale Retinex algorithm.

[0027] In one implementation method, the principle of adaptive gamma correction is as follows: For each pixel (x, y) in the image, take a local window of size w×w centered on it and calculate the local mean; after normalizing the local mean, perform local brightness adaptive adjustment to obtain the gamma value; The corresponding pixel value is calculated based on each pixel and its corresponding gamma value; all pixel values ​​are combined to obtain the first enhanced mold spot image; The formula for calculating gamma value is: in, This represents the pixel value after gamma correction. This represents the gamma value of pixel (a, b). This represents the local mean of pixel (a, b). This represents the local mean of pixel (a, b) after normalization. denoted by , w represents the preset size of the local window, f(x,y) represents the input enhanced mold image, a and b represent pixel coordinates, and i and j represent offset coordinates.

[0028] In one implementation, image enhancement is performed on the dried fruit mold image. Specifically, by enhancing the edge details of the mold spots, the corresponding sugar frosting image will be weakened, thereby enhancing the contrast of the mold spot image. Then, local texture enhancement is performed on the mold spot image to improve the accuracy of subsequent mold spot recognition.

[0029] In one implementation, see [link to implementation details]. Figure 2 , Figure 2 This is a network structure diagram of an anomaly recognition model provided in an embodiment of the present invention. See [link / reference]. Figure 3 , Figure 3 This invention provides a network structure diagram of a YOLOv8 model. The anomaly detection model is obtained by modifying the YOLOv8 model, specifically including: In the main structure and neck structure, the connection between the SPPF module and the Upsample module is disconnected, and an enhanced feature extraction module is added after the SPPF module and connected to the Upsample module; in the detection head, the original three Detect modules are replaced with the Detect_1 module and connected to the neck structure C2f module. The working principle of the enhanced feature extraction module is as follows: The input to the enhanced feature extraction module is determined as the input feature map; The input feature map is subjected to horizontal average pooling and vertical average pooling to obtain the horizontal feature map and the vertical feature map, respectively. The horizontal feature map is segmented to obtain multiple sub-horizontal feature maps; The vertical feature map is segmented to obtain multiple sub-vertical feature maps; The first feature map is obtained by convolving multiple sub-horizontal feature maps and then concatenating them; the second feature map is obtained by convolving multiple sub-vertical feature maps and then concatenating them. The first and second feature maps are normalized and then concatenated to obtain the third feature map. The fourth feature map is obtained by performing a depthwise convolution on the third feature map; The fourth feature is normalized to obtain the fifth feature map; the third and fifth feature maps are multiplied element by element to obtain the target feature map.

[0030] In one implementation, specifically in a particular embodiment, the horizontal feature map is segmented to obtain multiple sub-horizontal feature maps, and the vertical feature map is segmented to obtain multiple sub-vertical feature maps, where the number of sub-feature maps is N (4, 8, etc.). An enhanced feature extraction module is used to improve the efficiency of extracting mold spots on dried fruit features, and the Detect_1 module improves the processing speed and detection efficiency of dried fruit images.

[0031] In one implementation, the working principle of the Detect_1 module specifically includes: The three output features of the neck structure are respectively defined as three input features; the three input features are the first feature, the second feature, and the third feature. The first feature is upsampled to obtain the fourth feature, and the fourth feature is convolved to obtain the first convolutional feature; The first fused feature is obtained by fusing the second feature and the first convolutional feature. The second feature is upsampled to obtain the fifth feature, and the fifth feature is convolved to obtain the second convolutional feature; The second fused feature is obtained by fusing the third feature and the second convolutional feature. The first fusion feature is upsampled and then concatenated with the second fusion feature to obtain the concatenated feature. The concatenated feature is then substituted into the Detect module to obtain the recognition result.

[0032] In one implementation, the mold spot detection module on the surface of dried fruit utilizes a hierarchical upsampling, convolution, and feature fusion approach for the three-level features output from the neck structure, fully leveraging feature information at different scales and levels. By upsampling and convolving low-level features and fusing them with mid-level features, and upsampling and convolving mid-level features and fusing them with high-level features, the detailed features and semantic information of small targets like mold spots are enhanced, effectively avoiding missed or false detections of mold spots caused by single features. The two-level fused features are then upsampled and concatenated before being fed into the detection module, which can fully integrate multi-scale contextual information, improving the ability to identify small mold spots and localized moldy areas on the dried fruit surface. While ensuring real-time detection, this significantly improves the efficiency and accuracy of mold spot detection.

[0033] Based on the same inventive concept, this invention also provides a dried fruit cold storage and preservation control system. See also Figure 4 , Figure 4 A framework diagram of a dried fruit cold storage and preservation control system provided in an embodiment of the present invention includes: Data acquisition module: Acquires the surface humidity value of dried fruit immediately after it is placed in the cold storage and the humidity range of the cold storage; the humidity range includes the lower limit and the upper limit of the range; Humidity control module: It compares the humidity value of the dried fruit surface with the humidity range of the cold storage; if the humidity value of the dried fruit surface is greater than the upper limit of the range, it will dehumidify the dried fruit surface to the humidity range; if the humidity value of the dried fruit surface is less than the upper limit of the range, it will moisturize the dried fruit surface to the humidity range. Image acquisition module: Real-time monitoring of dried fruit placed in cold storage to obtain images of dried fruit; Preprocessing module: preprocesses the dried fruit images to obtain dried fruit mold image and dried fruit sugar frosting image; Image enhancement module: Enhances the image of mold spots on dried fruit to obtain the target dried fruit image; Mold Spot Recognition Module: Substitutes the target dried fruit image into the anomaly recognition model to obtain the recognition result; the recognition result is the location and size of the mold spots on the dried fruit. Backend processing module: Marks the dried fruit with mold spots in the recognition results and transmits them to the backend for processing.

[0034] The dried fruit cold storage and preservation control system provided by this invention controls the humidity of dried fruit within the humidity range of the cold storage, reduces water droplet condensation, decreases the occurrence of mold spots, and improves the preservation quality of dried fruit. By comparing and differentiating images of mold spots and sugar frosting on dried fruit, and by performing image enhancement on the images of mold spots, the accuracy of mold spot identification is improved. The accuracy of mold spot identification is also improved through an anomaly identification model.

[0035] 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 method for controlling the cold storage and preservation of dried fruit, characterized in that, The method includes: Obtain the surface humidity value of the dried fruit immediately after it is placed in the cold storage and the humidity value range of the cold storage; the humidity value range includes a lower limit and an upper limit. The humidity value of the dried fruit surface is compared with the humidity range of the cold storage. If the humidity value of the dried fruit surface is greater than the upper limit of the range, the dried fruit surface is dehumidified until it is within the humidity range. If the humidity value of the dried fruit surface is less than the upper limit of the range, the dried fruit surface is moisturized until it is within the humidity range. Then, real-time monitoring of the dried fruit placed in the cold storage is used to obtain images of the dried fruit; The dried fruit images are preprocessed to obtain dried fruit mold images and dried fruit sugar frosting images; Image enhancement is performed on images of mold spots on dried fruit to obtain the target image of dried fruit; The target dried fruit image is substituted into the anomaly recognition model to obtain the recognition result; the recognition result is the location and size of the mold spots on the dried fruit. The dried fruit with mold spots in the identification results are marked and transmitted to the backend for processing.

2. The method for controlling the cold storage and preservation of dried fruit according to claim 1, characterized in that, Preprocessing the dried fruit images to obtain dried fruit mold image and dried fruit frosting image specifically includes: Convert the dried fruit image from RGB format to HSV format; The pixel values ​​of the H and S channels in the converted dried image are extracted and compared with the preset mold color threshold range and sugar frost color threshold range, respectively. Pixel regions that meet the mold color threshold range are divided into initial mold regions, and pixel regions that meet the frosting color threshold range are divided into initial frosting regions. Morphological opening operations were performed on the initial moldy area and the initial sugar frosting area to obtain the dried fruit moldy image and the dried fruit sugar frosting image, respectively.

3. The method for controlling the cold storage and preservation of dried fruit according to claim 2, characterized in that, The specific formula for the morphological opening operation includes: in, Represents the image after morphological opening operation. The binary image obtained from the initial segmentation is represented by a pixel value of 0 or 1, where 1 represents a candidate region and 0 represents a background region; E represents the structuring element, x and y represent pixel coordinates, r represents the radius of the structuring element in pixels, and is used to define the shape and size of the neighborhood in morphological operations; m and n represent the offset coordinates in the structuring element, where m is the offset in the row direction and n is the offset in the column direction. This represents the erosion operator. This represents the dilation operator, where i and j represent the pixel coordinates of the current image.

4. The method for controlling the cold storage and preservation of dried fruit according to claim 1, characterized in that, The process of enhancing the image of mold spots on dried fruit to obtain the target dried fruit image includes: The enhanced mold image is subjected to adaptive gamma correction to obtain a first enhanced mold image; the second enhanced mold image is an image with enhanced mold edge details. The target dried fruit image is obtained by calculating the first enhanced moldy image using the multi-scale Retinex algorithm.

5. The method for controlling the cold storage and preservation of dried fruit according to claim 4, characterized in that, The principle and process of the adaptive gamma correction include: For each pixel in the image, a local window of a preset size is taken with the pixel as the center, and the local mean is calculated; after normalizing the local mean, the local brightness is adaptively adjusted to obtain the gamma value; The corresponding pixel value is calculated based on each pixel and its corresponding gamma value; all pixel values ​​are combined to obtain the first enhanced mold image.

6. The method for controlling the cold storage and preservation of dried fruit according to claim 5, characterized in that, The formula for calculating the gamma value specifically includes: in, This represents the pixel value after gamma correction. This represents the gamma value of pixel (a, b). This represents the local mean of pixel (a, b). This represents the local mean of pixel (a, b) after normalization. denoted by , w represents the preset size of the local window, f(x,y) represents the input enhanced mold image, a and b represent pixel coordinates, and i and j represent offset coordinates.

7. The method for controlling the cold storage and preservation of dried fruit according to claim 1, characterized in that, The anomaly detection model is derived from modifications of the YOLOv8 model, specifically including: In the main structure and neck structure, the connection between the SPPF module and the Upsample module is disconnected, and an enhanced feature extraction module is added after the SPPF module and connected to the Upsample module. In the detection head, the original three Detect modules are replaced with the Detect_1 module, which is then connected to the neck structure C2f module.

8. The method for controlling the cold storage and preservation of dried fruit according to claim 7, characterized in that, The working principle of the enhanced feature extraction module specifically includes: The input to the enhanced feature extraction module is determined as the input feature map; The input feature map is subjected to horizontal average pooling and vertical average pooling to obtain a horizontal feature map and a vertical feature map, respectively. The horizontal feature map is segmented to obtain multiple sub-horizontal feature maps; The vertical feature map is segmented to obtain multiple sub-vertical feature maps; The first feature map is obtained by convolving the multiple sub-horizontal feature maps and then concatenating them; the second feature map is obtained by convolving the multiple sub-vertical feature maps and then concatenating them. The first and second feature maps are normalized and then concatenated to obtain the third feature map. The third feature map is subjected to depthwise convolution to obtain the fourth feature map; The fourth feature is normalized to obtain the fifth feature map; the third feature map and the fifth feature map are multiplied element by element to obtain the target feature map.

9. The method for controlling the cold storage and preservation of dried fruit according to claim 7, characterized in that, The working principle of the Detect_1 module specifically includes: The three output features of the neck structure are respectively determined as three input features; the three input features are the first feature, the second feature, and the third feature; The first feature is upsampled to obtain the fourth feature, and the fourth feature is convolved to obtain the first convolutional feature; The second feature and the first convolutional feature are fused to obtain the first fused feature; The second feature is upsampled to obtain the fifth feature, and the fifth feature is convolved to obtain the second convolutional feature; The third feature and the second convolutional feature are fused to obtain the second fused feature; The first fusion feature is upsampled and then concatenated with the second fusion feature to obtain the concatenated feature. The concatenated feature is then substituted into the Detect module to obtain the recognition result.

10. A dried fruit cold storage and preservation control system, used to implement the dried fruit cold storage and preservation control method according to any one of claims 1-9, characterized in that, The system includes: Data acquisition module: acquires the surface humidity value of dried fruit immediately after it is placed in the cold storage and the humidity value range of the cold storage; the humidity value range includes a lower limit and an upper limit. Humidity control module: compares the surface humidity value of the dried fruit with the humidity range of the cold storage; if the surface humidity value of the dried fruit is greater than the upper limit of the range, the surface of the dried fruit is dehumidified until it is within the humidity range; if the surface humidity value of the dried fruit is less than the upper limit of the range, the surface of the dried fruit is moisturized until it is within the humidity range. Image acquisition module: Real-time monitoring of dried fruit placed in cold storage to obtain images of dried fruit; Preprocessing module: preprocesses the dried fruit image to obtain dried fruit mold image and dried fruit sugar frosting image; Image enhancement module: Enhances the image of mold spots on dried fruit to obtain the target dried fruit image; Mold Spot Recognition Module: Substitutes the target dried fruit image into the anomaly recognition model to obtain the recognition result; the recognition result is the location and size of the mold spots on the dried fruit. Backend processing module: Marks the dried fruit with mold spots in the recognition results and transmits them to the backend for processing.

Images