Intelligent management method for breeding of medicinal stiff silkworms

Abstract

The invention discloses an intelligent management method for medicinal bombyx batryticatus culture, and belongs to the technical field of intelligent management, and the method comprises the steps: placing silkworm chrysalis inoculated with beauveria bassiana in the same batch at a specimen collection point, and obtaining a monitoring data sequence, a monitoring image set and inoculation parameters of the beauveria bassiana corresponding to the specimen collection point in real time; identifying the silkworm chrysalis by the image identification model, outputting a silkworm chrysalis stiffness degree sequence, and calculating a silkworm chrysalis stiffness period; inputting all the sequence sets into a monitoring model for analysis, generating a state management model, generating a state estimation table corresponding to the specimen collection points, and constructing a state condition list of silkworm chrysalis; preprocessing all collected data of the silkworm chrysalis to be monitored, inputting the preprocessed data into the state management model, outputting a stiffness state, judging whether the stiffness state is qualified or not based on the state condition list, and repeating the step until the stiffness process of the silkworm chrysalis to be monitored is completed to form the medicinal stiff silkworm. According to the invention, the intelligence and accuracy of the breeding management of the medicinal stiff silkworms can be improved.

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent management, and in particular relates to an intelligent management method for breeding medicinal silkworms. Background Art

[0002] As a biological resource with important medicinal value, the cultivation and collection of medicinal silkworm has always been an important part of the traditional Chinese medicine industry. Traditionally, the cultivation of silkworms mainly depends on the natural environment and artificial experience, lacking systematic scientific management, resulting in low breeding efficiency, unstable product quality, and being easily affected by climate and environmental changes, making it difficult to achieve precise regulation, thus affecting the growth quality and yield of silkworms. In order to improve the breeding efficiency and product quality of medicinal silkworms, it is necessary to introduce modern information technology, such as the Internet of Things, big data, and artificial intelligence, to achieve intelligent management of the breeding process, thereby improving resource utilization, reducing costs, and enhancing industrial competitiveness, achieving precise control of the breeding environment, optimizing disease prevention and control strategies, and realizing automated data collection and analysis, and ultimately improving the overall efficiency and product quality of silkworm breeding to meet the market demand for high-quality medicinal silkworms.

[0003] Similar prior art includes a Chinese patent application with publication number CN118710176A, which discloses a smart dry cocoon storage environment management system, which includes: regularly obtaining the cocoon pattern feedback from the sampling inspection system, and identifying whether the cocoon has an abnormality based on the cocoon pattern, and in the case of an abnormality, identifying the abnormality type corresponding to the cocoon pattern; adjusting the local environmental parameters in the corresponding dry cocoon storage based on the abnormality type, thereby slowing down the deterioration trend of the corresponding abnormality, and the sampling inspection system is used to regularly obtain the dry cocoon preservation patterns of different areas in the dry cocoon storage warehouse, and employees regularly place the dry cocoons in different areas of the dry cocoon storage warehouse according to the corresponding positions on the sampling inspection table, so that the sampling inspection system generates the corresponding cocoon pattern obtained by the camera, and then feeds the cocoon pattern back to the dry cocoon storage environment management system, so that it can automatically adjust the local environmental parameters in the dry cocoon storage in time according to the abnormal situation reflected by the cocoon pattern. There is also a Chinese patent application with publication number CN118333462A, which provides a modern silkworm industrial park digital platform and management method, which involves the field of breeding management technology. The invented platform is centered around information collection terminals, radio frequency identification terminals, and cloud-based digital management units, and realizes real-time monitoring and management of various functional areas. Silkworm breeding, strain cultivation, spraying inoculation, silkworm cultivation, finished product screening, and quality analysis all have special management modules to monitor and update various status information to respond to possible abnormal situations. This comprehensive digital management system improves the transparency, efficiency, and quality of the production process, reduces production risks and losses, and enhances the competitiveness and sustainable development capabilities of the industrial park. Through the use of scientific and technological means, the platform has brought new possibilities for modern breeding management, helping the industry move towards a more intelligent, efficient, and controllable direction.

[0004] However, the above-mentioned existing technologies cannot automatically identify the abnormal types corresponding to the cocoon images, the digital platform requires more complex equipment, and cannot timely monitor the abnormal types of silkworm breeding. In actual situations, due to the different degrees of influence of conditional variables, the rigidification state of silkworm pupae will also be different. It is necessary to use conditional probability to improve the accuracy of medicinal silkworm breeding management. Summary of the invention

[0005] In view of the above-mentioned technical problems, the present invention provides an intelligent management method for breeding medicinal silkworms to solve the problems in the prior art.

[0006] The present invention provides an intelligent management method for breeding medicinal silkworms, comprising:

[0007] A plurality of specimen collection points are set in the breeding area, and silkworm pupae inoculated with Beauveria bassiana from the same batch are placed at the specimen collection points, and the sensor device combination acquires the monitoring data sequence, monitoring image set and Beauveria bassiana inoculation parameters corresponding to the specimen collection points in real time based on a preset collection interval;

[0008] Constructing an image recognition model, wherein the image recognition model recognizes silkworm pupae based on the monitoring image set, generates a classified and annotated image set, outputs a sequence of rigidity of silkworm pupae based on the classified and annotated image set, calculates the rigidity period of silkworm pupae based on the rigidity sequence, and combines the monitoring data sequence, the rigidity sequence and the classified and annotated image set into a sequence set based on the rigidity period;

[0009] Constructing a monitoring model based on a Bayesian network, inputting all the sequence sets into the monitoring model for analysis based on the collection interval, the sensor device combination and the inoculation parameters, generating a state management model, generating a state estimation table corresponding to the specimen collection point, and constructing a state condition list of silkworm pupae based on all the state estimation tables;

[0010] Acquire all the collected data corresponding to the current collection time period of the silkworm pupae to be monitored, pre-process the collected data and input them into the state management model, output the rigidification state of the silkworm pupae to be monitored, and judge whether the rigidification state is qualified based on the state condition list. If so, analyze all the collected data corresponding to the next collection time period in real time. Otherwise, output an unqualified condition type, and adjust the control device corresponding to the sensor device combination based on the unqualified condition type. Repeat this step until the silkworm pupae to be monitored complete the rigidification process to form medicinal silkworm pupae.

[0011] Furthermore, the image recognition model identifies silkworm pupae based on the monitoring image set, comprising the following steps:

[0012] Installing a camera device just above the specimen collection point, the camera device generates monitoring images of silkworm pupae by shooting based on preset shooting parameters and the collection interval, sequentially combining all the monitoring images to generate the monitoring image set, extracting a plurality of the monitoring images from the monitoring image set as initial images, and combining them to generate the initial image set;

[0013] Converting the initial image into a grayscale image based on the shooting parameters, respectively obtaining the RGB value and pixel coordinates of any pixel in the grayscale image, constructing a three-dimensional coordinate system based on the RGB value, plotting any pixel in the grayscale image in the three-dimensional coordinate system based on the RGB value to generate a spatial point, clustering all the spatial points based on spatial distribution positions using a uniform clustering algorithm to generate multiple cluster clusters, calculating a first distance between any spatial point in the cluster cluster and a cluster center point, and calculating an average value and a standard deviation of all the first distances;

[0014] Mapping all the spatial points contained in the clusters to the grayscale image corresponding to the initial image based on the pixel coordinates, and performing category marking on the pixels corresponding to all the clusters to generate category labels;

[0015] The image recognition model is constructed based on a convolutional neural network. The image recognition model is trained on the initial image containing the category label and the remaining monitoring images, and the category labels contained in all the monitoring images are identified based on the recognition scores. The pixels in the monitoring images whose category labels are silkworm pupa labels are identified as silkworm pupa, the monitoring images with the category labels are set as classified annotated images, and all the classified annotated images are combined based on the acquisition interval and set as the classified annotated image set.

[0016] Furthermore, the step of outputting the rigidity degree sequence of silkworm pupae based on the classified and labeled image set comprises the following steps:

[0017] Obtain pixels identified as silkworm pupae in the classified and labeled image and set them as monitoring pixels, connect adjacent monitoring pixels in sequence based on the pixel coordinates to generate a monitoring area, and set the total number of pixels included in the monitoring area as a first value;

[0018] Based on the acquisition interval, a center point pixel corresponding to the monitoring area included in the first classified and labeled image in the classified and labeled image set is obtained, and the coordinates of the center point pixel are obtained and set as fixed coordinates, and a second distance between an edge pixel of the monitoring area and the fixed coordinates is obtained based on the pixel coordinates, and an average value corresponding to all the second distances in the monitoring area is calculated and set as a second value;

[0019] Taking the fixed coordinate as a constant point, calculating the monitoring area corresponding to the constant point in the remaining classified and labeled images in turn to generate the first value and the second value, accumulating all the first values ​​in the classified and labeled images in turn to generate an area value, and setting the average value of all the second values ​​in the classified and labeled images as a growth value;

[0020] The area values ​​and the growth values ​​corresponding to all the classified and labeled images are counted in sequence based on the acquisition interval to generate a first area sequence and a first growth sequence, and the first area sequence and the first growth sequence are combined and set as the rigidity degree sequence of silkworm pupae.

[0021] Furthermore, the step of identifying the category labels contained in all the monitoring images based on the recognition scores comprises the following steps:

[0022] Calculate the recognition score F between any pixel in the grayscale image and the category label based on a first formula, wherein the first formula is: Wherein, D is the first distance of the spatial point corresponding to any pixel in the grayscale image, α is the average value of all the first distances in the cluster, and β is the standard deviation of all the first distances;

[0023] If the recognition score is less than or equal to the first threshold, it is determined that the classification of the spatial points contained in the cluster is accurate, and any pixel in the grayscale image belongs to the same category label as the cluster; otherwise, the parameters of the uniform clustering algorithm are adjusted and all the spatial points are re-clustered until the recognition score is less than or equal to the first threshold.

[0024] Further, the calculating of the rigidification period of silkworm pupae based on the rigidification degree sequence comprises:

[0025] The rigidity degree sequence is fitted to generate a first type of curve, an inflection point of the first type of curve is marked, and the first type of curve is divided into a plurality of rigid extension areas and rigid drying areas based on the inflection point, and a time period included in the rigid extension area and the next adjacent rigid drying area is obtained, and the time period is set as the rigidity period.

[0026] Furthermore, the state management model is generated based on the following steps:

[0027] The monitoring model extracts multiple conditional variables based on the data information contained in the sequence set, performs feature selection on the data information based on the data type to generate a feature quantity set, sets the conditional variables as child nodes, calculates the conditional probability between any of the child nodes and a preset parent node based on the feature quantity set, and performs modeling based on the conditional probability, the preset parent node and the child node to generate the state management model.

[0028] Further, a status condition list of silkworm pupae is constructed based on the following steps:

[0029] Respectively obtaining the category label and the total number of silkworm pupae with the category label in the classified and labeled image and setting them as a third value and a fourth value, setting a plurality of growth state types based on the ratio between the fourth value and the third value, and setting the growth state type as the preset parent node;

[0030] Acquire the state information of the child node, summarize the first occurrence number of any of the preset parent nodes and the second occurrence number of the same state information corresponding to any of the preset parent nodes, and set the ratio between the second occurrence number and the first occurrence number as the conditional probability between the child node and the preset parent node;

[0031] Based on the time feature, the preset parent node, the child node and the conditional probability are combined to generate the state estimation table, all the state estimation tables are summarized to generate a first data table, the conditional probabilities corresponding to the same preset parent node and the same child node are averaged and summarized to generate a second data table, and the preset parent node in the normal rigid state is extracted from the second data table and combined into a state condition list of silkworm pupae.

[0032] Further, judging whether the rigid state is qualified based on the state condition list includes the following steps:

[0033] The collected data includes a data sequence corresponding to the sensor device combination, a rigid state corresponding to the collected data is obtained based on the image recognition model and the state management model, and list information corresponding to the rigid state is checked in the second data table based on the collection time period;

[0034] If the list information belongs to the normal rigid state of the state condition list, the rigid state is determined to be qualified; otherwise, different sub-node items of the list information and the normal rigid state are screened out based on the current collection time period, and the condition variables corresponding to the different sub-node items are set to the unqualified condition type.

[0035] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0036] The present invention firstly collects data from each specimen collection point by setting a combination of multiple sensor devices, thereby improving the precision of the specimen collection points; then, classifies and identifies the monitoring image set through an image recognition model, and generates a sequence of the rigidity of silkworm pupae, thereby improving the accuracy of identifying silkworm pupae in the monitoring images; finally, a monitoring model is constructed through a Bayesian network, thereby accurately analyzing the conditional probability of silkworm pupae in the rigidity breeding process, thereby improving the accuracy of intelligent management.

[0037] The present invention also combines the image recognition model and the state management model to analyze the collected data of the monitored silkworm pupae in the current collection time period, outputs the rigidification state, and checks the eligibility of the rigidification state in combination with the state condition list to adjust the control equipment, thereby improving the intelligence of medicinal silkworm breeding. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flowchart of the steps of an intelligent management method for medicinal silkworm breeding according to the present invention;

[0039] Figure 2 is a schematic diagram of the first type of curve in the present invention;

[0040] Figure 3 It is a schematic diagram of the state management model in the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0042] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0043] like Figure 1 As shown, an intelligent management method for breeding medicinal silkworms comprises:

[0044] S1: Multiple specimen collection points are set up in the breeding area, and silkworm pupae inoculated with Beauveria bassiana from the same batch are placed at the specimen collection points. The sensor device combination obtains the monitoring data sequence, monitoring image set and Beauveria bassiana inoculation parameters corresponding to the specimen collection points in real time based on the preset collection interval.

[0045] Specifically, in the present embodiment, the breeding area refers to the indoor production space, the specimen collection point refers to the location area in the breeding area for placing the silkworm pupa device, wherein the number of the specimen collection points may be multiple. The same batch refers to the batch number of silkworm pupae inoculated with Beauveria bassiana at the same time, wherein silkworm pupae are a general term, and therefore, the number of silkworm pupae is multiple. The sensor device combination refers to a collection of equipment types used to collect a variety of environmental data, including but not limited to humidity sensors, temperature sensors, cameras and other equipment types. The acquisition interval refers to the time information composed of acquisition time points composed of the same time interval, the monitoring data sequence refers to the data sequence of the non-image sensor device combination monitoring and collecting the specimen collection points in sequence by the acquisition interval, the monitoring image set refers to the collection of monitoring images generated by the camera shooting the specimen collection points in sequence by the acquisition interval, and the inoculation parameters of Beauveria bassiana refer to the performance values ​​of the inoculated Beauveria bassiana, including but not limited to spore yield, enzyme activity, bioconversion efficiency, etc.

[0046] S2: Construct an image recognition model. The image recognition model identifies silkworm pupae based on the monitoring image set, generates a classified and labeled image set, and outputs a rigidity degree sequence of silkworm pupae based on the classified and labeled image set. The rigidity period of silkworm pupae is calculated based on the rigidity degree sequence, and the monitoring data sequence, the rigidity degree sequence and the classified and labeled image set are combined into a sequence set based on the rigidity period.

[0047] Specifically, in the present embodiment, the image recognition model is a model constructed by a convolutional neural network for image classification and image recognition of a monitoring image set. Identifying silkworm pupae refers to identifying pixel areas with category labels of silkworm pupae in the monitoring image set, and the classified and labeled image set refers to a collection of classified and labeled images containing various category label information. The rigidity degree sequence refers to a data sequence of state changes of silkworm pupae generated by analyzing the classified and labeled image set. The rigidity period refers to the length of time required for silkworm pupae to form medicinal silkworm pupae after being inoculated with Beauveria bassiana. The sequence set refers to a data set generated by combining the monitoring data sequence, rigidity degree sequence and classified and labeled images in the classified and labeled image set corresponding to the same rigidity period through the acquisition interval.

[0048] S3: Construct a monitoring model based on the Bayesian network, input all sequence sets into the monitoring model for analysis based on the collection interval, sensor equipment combination and inoculation parameters, generate a state management model, and generate a state estimation table corresponding to the specimen collection point, and construct a list of state conditions of silkworm pupae based on all state estimation tables.

[0049] Specifically, in the present embodiment, the Bayesian network is a probabilistic graph model for representing the conditional dependency between variables. It combines probability theory and graph theory, and displays the probabilistic dependency structure between variables in a graphical manner, which is convenient for probabilistic reasoning and decision analysis. The monitoring model constructed by the Bayesian network can analyze the state monitoring between the rigid state of silkworm pupae after inoculation with Beauveria bassiana and the environmental data corresponding to the sensor device combination and the Beauveria bassiana inoculation parameters. The sequence set corresponding to the same rigid cycle is sequentially input into the monitoring model based on the time information corresponding to the acquisition interval, and the analyzed monitoring model can be set as a state management model. The state estimation table refers to a list of relevant information for describing the various different state stages of silkworm pupae in the actual rigid process, and the state condition list refers to a list of relevant information for describing the various different stages of silkworm pupae in the ideal rigid process generated by summarizing all state estimation tables. Therefore, the difference between the state estimation table and the state condition list is that the ideal rigid process is generated by the actual rigid process, which can improve the accuracy of medicinal silkworm breeding.

[0050] S4: Obtain all collected data corresponding to the current collection time period of the silkworm pupae to be monitored, pre-process the collected data and input them into the state management model, output the rigidification state of the silkworm pupae to be monitored, and determine whether the rigidification state is qualified based on the state condition list. If so, analyze all collected data corresponding to the next collection time period in real time. Otherwise, output an unqualified condition type, and adjust the control device corresponding to the sensor device combination based on the unqualified condition type. Repeat this step until the silkworm pupae to be monitored complete the rigidification process to form medicinal silkworm pupae.

[0051] Specifically, in this embodiment, the collected data has the same source as the above-mentioned monitoring data sequence, monitoring image set and inoculation parameter, and the same collection interval, but the time point of collection is different, and the objects of collection are different, wherein the current collection time period refers to the time period that currently needs to be intelligently managed, that is, the collected data includes the monitoring data sequence, monitoring image set and Beauveria bassiana inoculation parameters corresponding to the silkworm pupae to be monitored in the current collection time period. Preprocessing refers to processing the collected data through the same steps as above, for example, using an image recognition model to perform image classification and image recognition on the monitoring image set in the collected data.

[0052] The silkworm pupae to be monitored refer to a new batch of silkworm pupae inoculated with Beauveria bassiana and the number is multiple. The rigid state refers to the state distribution of the morphological changes of the silkworm pupae to be monitored after being inoculated with Beauveria bassiana. Whether it is qualified is the basis for determining whether it is necessary to perform intelligent regulation and management on the silkworm pupae to be monitored. If it is qualified, it means that the silkworm pupae to be monitored in the current collection time period are in a normal rigid state, and only all the collected data corresponding to the next collection time period need to be analyzed in real time. If it is unqualified, it means that there are influencing factors in the rigid state of the silkworm pupae to be monitored in the current collection time period, that is, unqualified condition types. Therefore, it is necessary to adjust the control equipment or inoculation parameters corresponding to the condition type through the state condition list. Repeat this step to monitor and analyze the silkworm pupae to be monitored until the silkworm pupae to be monitored complete the entire rigid process to form a medicinal silkworm. For example, the rigid state of the silkworm pupae group A to be monitored is "rigidity delay state". Through the state condition list, it can be analyzed that the rigid state is unqualified, and the unqualified condition type is "temperature is less than 30", so the control equipment used to adjust the temperature is controlled to adjust the temperature to 30.

[0053] As a preferred technical solution of the present invention, the image recognition model identifies silkworm pupae based on the monitoring image set, comprising the following steps:

[0054] A camera device is installed directly above the specimen collection point. The camera device generates monitoring images of silkworm pupae based on preset shooting parameters and collection intervals. All monitoring images are combined in sequence to generate a monitoring image set. Multiple monitoring images are extracted from the monitoring image set as initial images, and the initial image set is generated by combining them.

[0055] The initial image is converted into a grayscale image based on the shooting parameters, the RGB value and pixel coordinates of any pixel in the grayscale image are obtained respectively, a three-dimensional coordinate system is constructed based on the RGB value, any pixel in the grayscale image is plotted in the three-dimensional coordinate system based on the RGB value to generate a spatial point, all spatial points are clustered based on the spatial distribution position using a uniform clustering algorithm to generate multiple cluster clusters, the first distance between any spatial point in the cluster and the cluster center point is calculated, and the average and standard deviation of all first distances are calculated.

[0056] All spatial points contained in the clusters are mapped to the grayscale image corresponding to the initial image based on the pixel coordinates, and the pixels corresponding to all clusters are marked with categories to generate category labels.

[0057] An image recognition model is constructed based on a convolutional neural network. The image recognition model is trained on the initial image containing category labels and the remaining monitoring images, and the category labels contained in all monitoring images are identified based on the recognition scores. The pixels in the monitoring images with category labels as silkworm pupa labels are identified as silkworm pupae, and the monitoring images with category labels are set as classified annotated images. All classified annotated images are combined based on the acquisition interval and set as a classified annotated image set.

[0058] Specifically, in this embodiment, the camera device refers to the device that is a camera in the sensor device combination, and the shooting parameters refer to the imaging performance parameters of the camera device and the vertical distance at which the camera device is installed directly above the specimen collection point. The initial image refers to a monitoring image randomly selected from the monitoring image set, and the initial image set refers to a collection of multiple initial images selected.

[0059] The brightness and contrast of the initial image can be controlled by adjusting the shooting parameters such as exposure, ISO, and aperture, and then grayscale conversion is performed to generate a grayscale image. RGB values ​​include R values, G values, and B values. The R values, G values, and B values ​​of any pixel in a grayscale image are equal, that is, the RGB value is equal to the grayscale value. Pixel coordinates refer to the row and column coordinate values ​​corresponding to any pixel in a grayscale image. The three-dimensional coordinate system refers to the spatial coordinates in which the x, y, and z axes represent the R value, G value, and B value, respectively. A spatial point refers to the point position generated by drawing each pixel in a grayscale image in a three-dimensional coordinate system according to the RGB value. The uniform clustering algorithm generally refers to a clustering method that divides spatial points into multiple clusters, where the points in each cluster are close to each other and relatively far away from the points in other clusters, such as the K-Means algorithm, hierarchical clustering, and other algorithms. Spatial distribution refers to the coordinate distribution of spatial points in a three-dimensional coordinate system. Clustering clusters refer to the results of the uniform clustering algorithm, so there are multiple clustering clusters. The first distance refers to the spatial distance between a spatial point and a cluster center point in a three-dimensional coordinate system, wherein the cluster center point refers to the spatial point located at the center of a cluster, the average value refers to the average value of the first distances corresponding to all spatial points contained in the cluster, and the standard deviation refers to the degree of dispersion of the average values ​​of the first distances corresponding to all spatial points contained in the cluster, wherein the average value and the standard deviation can be used later to evaluate the clustering effect of the cluster, that is, the identification score.

[0060] The pixel corresponding to each spatial point has pixel coordinates. Therefore, mapping refers to inverting all spatial points contained in any cluster to the initial image through pixel coordinates, which can form the category distribution of the same cluster in the initial image, and mark the same cluster with a category label in the initial image, wherein the category label refers to the label information of the material category in the initial image, including but not limited to silkworm pupae, insect pests, plants, background and other labels. Compared with directly using the edge segmentation of the image for image classification and recognition, this step can reduce the phenomenon that the edges are difficult to segment in the rigid breeding process of silkworm pupae with high density, thereby reducing the accuracy of classification and recognition. At the same time, this step can improve the accuracy of the category label in the initial image by manually marking the category.

[0061] Convolutional Neural Network (CNN) is a deep learning model that performs well in image recognition, video analysis, natural language processing and other fields. The core of CNN is that it can automatically and effectively learn spatial hierarchical features from data, which makes it particularly effective in processing image data with obvious spatial hierarchical structure. The initial image containing the category label and the remaining monitoring images with unlabeled category labels are combined into a training set and input into the image recognition model for training. The corresponding category labels in all monitoring images can be identified through semi-supervised learning. Semi-supervised learning is a machine learning paradigm that lies between fully supervised learning and fully unsupervised learning. In semi-supervised learning, the training of the image recognition model uses a part of labeled data (with labels) and a part of unlabeled data (without labels), which can easily obtain a large amount of unlabeled data, that is, the remaining monitoring images. Among them, the recognition score is an evaluation value of the clustering accuracy of the uniform clustering algorithm. The silkworm pupae in any monitoring image can be identified by the category label. At the same time, the monitoring image marked with the category label is set as the classified labeled image. All the classified labeled images can be combined based on the acquisition interval to set as the classified labeled image set.

[0062] As a preferred technical solution of the present invention, outputting a sequence of rigidity levels of silkworm pupae based on a classified and labeled image set comprises the following steps:

[0063] Pixels identified as silkworm pupae in the classified and labeled image are obtained and set as monitoring pixels, adjacent monitoring pixels are sequentially connected based on pixel coordinates to generate a monitoring area, and the total number of pixels included in the monitoring area is set as a first value.

[0064] Based on the acquisition interval, the center point pixel corresponding to the monitoring area contained in the first classified and labeled image is obtained in the classified and labeled image set, the coordinates of the center point pixel are obtained and set as fixed coordinates, the second distance between the edge pixel of the monitoring area and the fixed coordinates is obtained based on the pixel coordinates, and the average value corresponding to all the second distances in the monitoring area is calculated and set as the second value.

[0065] Taking the fixed coordinates as the constant points, the monitoring areas corresponding to the constant points in the remaining classified and labeled images are calculated in turn to generate the first value and the second value, all the first values ​​in the classified and labeled images are accumulated in turn to generate the area value, and the average value of all the second values ​​in the classified and labeled images is set as the growth value.

[0066] The area values ​​and growth values ​​corresponding to all the classified labeled images are counted in sequence based on the acquisition interval to generate a first area sequence and a first growth sequence, which are combined and set as a rigidity degree sequence of silkworm pupae.

[0067] Specifically, in this embodiment, in order to monitor the rigidification process of silkworm pupae, the change area of ​​the monitoring pixels can be analyzed. The monitoring pixels are sequentially extended to adjacent monitoring pixels to depict the range area of ​​the monitoring pixels, i.e., the monitoring area. Since there may be non-adjacent monitoring pixels in the classified and labeled image, there may be multiple monitoring areas in the classified and labeled image. The first value refers to the total number of pixels contained in any monitoring area.

[0068] The first classified annotated image refers to the arrangement sequence composed of the acquisition time corresponding to the acquisition interval in sequence. Therefore, the first classified annotated image represents the monitoring image corresponding to the initial stage of the silkworm pupa after inoculation with Beauveria bassiana. The center point pixel refers to the pixel located at the center point in the monitoring area (irregular), which can be directly obtained by computer vision technology, or calculated and generated by the pixel coordinates of the monitoring area. The fixed coordinates refer to the pixel coordinates of any center point pixel in the classified annotated image. Since there may be multiple monitoring areas in the same classified annotated image, the fixed coordinates may also be multiple. The edge pixel refers to the pixel located at the edge of the monitoring area, and the second distance refers to the straight-line distance between the pixel coordinates of the edge pixel and the fixed coordinate. In order to reduce the statistics of the second distance, the edge pixels can be screened by adjacent equal intervals. For example, among all the edge pixels of the monitoring area, pixels are extracted at equal intervals in sequence according to the total number of all edge pixels as edge pixels. The second value refers to the average value after the accumulation of all the second distances contained in any monitoring area, which can be used as the initial judgment value.

[0069] A constant point refers to a pixel that maintains a fixed coordinate unchanged. By the same method as above, the first and second values ​​of the monitoring area corresponding to the constant point are calculated in turn in the remaining classified and labeled images with the constant point as the center. The change sequence of the first value and the change sequence of the second value corresponding to the constant point can be obtained in turn according to the time information corresponding to the acquisition interval. Since there may be multiple constant points in any classified and labeled image, all the first values ​​are added together to evaluate the area occupied by the pixels belonging to the silkworm pupae in the classified and labeled image, that is, the area value. The average value of all the second values ​​can evaluate the change rate of the pixels belonging to the silkworm pupae in the classified and labeled image, that is, the growth value.

[0070] According to the time information corresponding to the acquisition interval, the area values ​​and growth values ​​corresponding to all the classified labeled images can be arranged and combined to generate a first area sequence and a first growth sequence. Among them, there should be a high correlation between the fluctuation trends corresponding to the first area sequence and the first growth sequence. Among them, the first area sequence and the first growth sequence can be used to determine whether the silkworm pupae meet the requirements during the rigidification process. Therefore, the first area sequence and the first growth sequence are combined into a rigidification degree sequence.

[0071] As a preferred technical solution of the present invention, identifying the category labels contained in all monitoring images based on the recognition scores includes the following steps:

[0072] The recognition score F between any pixel in the grayscale image and the category label is calculated based on the first formula, and the first formula is: Where D is the first distance of the spatial point corresponding to any pixel in the grayscale image, α is the average value of all first distances in the cluster, and β is the standard deviation of all first distances.

[0073] If the recognition score is less than or equal to the first threshold, it is determined that the spatial points contained in the cluster are accurately classified, and any pixel in the grayscale image belongs to the same category label as the cluster. Otherwise, adjust the parameters of the uniform clustering algorithm and re-cluster all spatial points until the recognition score is less than or equal to the first threshold.

[0074] Specifically, in this embodiment, the first formula is used to evaluate the accuracy of the clusters generated after clustering using the uniform clustering algorithm. The smaller the recognition score F, the more concentrated the spatial point corresponding to the pixel in the grayscale image is with the clustering cluster, and the better the clustering effect of the spatial point. Therefore, if the recognition score is less than or equal to the first threshold, it means that the spatial point corresponding to the pixel belongs to the clustering cluster, that is, the category label marked by the clustering cluster is used as the category label of the pixel in the grayscale image corresponding to the spatial point. Otherwise, it is necessary to adjust the parameters of the uniform clustering algorithm, such as the number of clusters, initial centroid selection, convergence parameters, distance measurement and other parameters. Until the recognition score between all pixels in the grayscale image and the corresponding category labels is less than or equal to the first threshold.

[0075] As a preferred technical solution of the present invention, calculating the rigidification period of silkworm pupae based on the rigidification degree sequence includes:

[0076] The rigidity degree sequence is fitted to generate a first type of curve, the inflection point of the first type of curve is marked, and the first type of curve is divided into multiple rigid extension areas and rigid drying areas based on the inflection point, and the time period included in the rigid extension area and the next adjacent rigid drying area is obtained, and the time period is set as the rigid period.

[0077] Specifically, in this embodiment, the first type of curve includes a fitting curve corresponding to the first area sequence in the rigidity degree sequence and a fitting curve corresponding to the first growth sequence, and a curve generated by the curvature change of the two fitting curves. The inflection point refers to the point where the curvature changes on the first type of curve, that is, the point where the concavity of the first type of curve changes. The rigid expansion area refers to the time period when the first type of curve rises slowly, and the rigid drying area refers to the time period when the first type of curve falls back. The first type of curve can be used to illustrate the changing rules of the rigidification process of silkworm pupae. The time period consisting of the starting time point of the rigid expansion area to the end time point of the next adjacent rigid drying area is the rigidification period. Figure 2 As shown, t1 to t2 are the starting time point and the ending time point of the rigid expansion area, t2 to t3 are the next adjacent rigid drying area of ​​the rigid expansion area, and the first type of curve after t4 is the curve corresponding to the next batch of silkworm pupae.

[0078] As a preferred technical solution of the present invention, a state management model is generated based on the following steps:

[0079] The monitoring model extracts multiple conditional variables based on the data information contained in the sequence set, selects features of the data information based on the data type to generate a feature set, sets the conditional variables as child nodes, calculates the conditional probability between any child node and the preset parent node based on the feature set, and builds a model based on the conditional probability, preset parent node and child node to generate a state management model.

[0080] Specifically, in the present invention, conditional variables refer to the types of sensor device combinations corresponding to data information with high correlation in a sequence set, such as temperature, humidity, inoculation parameters, etc. A feature quantity set refers to a set of feature quantities corresponding to data information of different data types, wherein a feature quantity refers to a numerical value extracted from data information that can represent or describe certain characteristics of the data, and feature selection is a process of extracting feature quantities of data information, such as principal component analysis, wavelet transform, and other feature selection methods.

[0081] The preset parent node refers to a characteristic quantity that is used to specify and is associated with a conditional variable. In the present invention, the growth state type of silkworm pupae is set as the preset parent node. The monitoring model generates the conditional probability between each child node and each preset parent node in a graphical manner by the Bayesian network, and then the Bayesian network performs statistical laws on the conditional probability, preset parent node, and child node to update the probability distribution of other variables in the network to establish a new model, namely, the state management model. Figure 3 As shown, in the constructed state management model, the preset parent nodes and child nodes contain various data information, among which K1, K2, K3, K4, K5, ..., Kn are all conditional probabilities between the corresponding preset parent nodes and child nodes.

[0082] As a preferred technical solution of the present invention, a list of silkworm chrysalis status conditions is constructed based on the following steps:

[0083] The category labels and the total number of silkworm pupae with the category labels in the classified annotation image are respectively obtained and set as the third value and the fourth value, a plurality of growth state types are set based on the ratio between the fourth value and the third value, and the growth state type is set as a preset parent node.

[0084] Obtain the state information of the child node, summarize the first occurrence number of any preset parent node and the second occurrence number of the same state information corresponding to any preset parent node, and set the ratio between the second occurrence number and the first occurrence number as the conditional probability between the child node and the preset parent node.

[0085] Based on the time characteristics, the preset parent nodes, child nodes and conditional probabilities are combined to generate a state estimation table, all state estimation tables are summarized to generate a first data table, the conditional probabilities corresponding to the same preset parent nodes and the same child nodes are averaged and summarized to generate a second data table, and the preset parent nodes of the normal rigid state are extracted from the second data table and combined into a state condition list of silkworm pupae.

[0086] Specifically, in this embodiment, the third value refers to the total number of pixels with category labels in the classified and annotated image, and the fourth value refers to the total number of pixels with category labels belonging to silkworm pupae in the classified and annotated image. Therefore, the ratio of the fourth value to the third value corresponding to all classified and annotated images is calculated in sequence, which can be used for subsequent identification and evaluation of the rigidification state of silkworm pupae. The growth state type refers to the morphological type information that may exist in the rigidification process of silkworm pupae, including but not limited to the plaque contamination state, the delayed rigidification state, the normal rigidification state, etc. Setting the information corresponding to the growth state type as a preset parent node can be conducive to building a complete Bayesian network.

[0087] State information refers to the data information of the characteristic quantity corresponding to the child node at different acquisition times. There are multiple state information. For example, the state information is the data information corresponding to "temperature 30, humidity 55, inoculation parameter M1, and rigidity degree data G1". Since the same preset parent node may correspond to the state information of multiple child nodes, the first number of occurrences refers to the number of state information of multiple child nodes corresponding to the same preset parent node, and the second number of occurrences refers to the number of state information of the same child node corresponding to the same preset parent node. The ratio of the second number of occurrences to the first number of occurrences is set as the conditional probability of the corresponding conditional variable of the child node.

[0088] The state estimation table refers to the tabular data generated based on the preset parent node, child node and conditional probability combination. Since there are multiple sample collection points, there are also multiple state estimation tables. The time feature refers to the time length information. Different time features may have the same preset parent node, but their corresponding child nodes may be different. For example, the preset parent node "normal rigid state" is in the extended state at the time length information T1, and in the dried state at the time length information T2, where the extended state and the dried state both belong to the "normal rigid state" of the preset parent node.

[0089] The state estimation tables corresponding to all specimen collection points are summarized to generate the first data table. Sum and average refers to a calculation method of first accumulating and then averaging. If the information corresponding to the preset parent node and the child node is the same, it means that the same silkworm pupa rigidification state exists at different sample collection points. Therefore, the corresponding conditional probabilities can be summed and averaged, and the first data table is summarized to generate the second data table. The normal rigid state is one of the preset parent nodes corresponding to the growth state type. Therefore, the table information corresponding to the normal rigid state is extracted in the second data table to generate a state condition list.

[0090] As a preferred technical solution of the present invention, judging whether the rigid state is qualified based on the state condition list includes the following steps:

[0091] The collected data includes a data sequence corresponding to the sensor device combination, the rigid state corresponding to the collected data is obtained based on the image recognition model and the state management model, and the list information corresponding to the rigid state is checked in the second data table based on the collection time period.

[0092] If the list information belongs to the normal rigid state of the state condition list, the rigid state is determined to be qualified. Otherwise, based on the current collection time period, different sub-node items of the list information and the normal rigid state are screened out, and the condition variables corresponding to the different sub-node items are set to unqualified condition types.

[0093] Specifically, in this embodiment, the collected data is preprocessed by the image recognition model and then input into the state management model for probability analysis, so that the rigid state corresponding to the current time period can be output, and then the data information corresponding to the rigid state of the current time period is selected from the second data table, that is, the list information.

[0094] By comparing the list information with the state condition list, it can be determined whether the rigid state is a normal rigid state. Different sub-node items refer to filtering out different data information in the sub-node according to the time length information corresponding to the current acquisition time period. For example, the sub-node corresponding to the normal rigid state is "temperature 35, humidity 70", while the sub-node corresponding to the rigid state is "temperature 35, humidity 50", so the data information corresponding to the humidity in the sub-node is a different sub-node item, and the condition variable corresponding to the data information corresponding to the humidity is set as the condition type.

[0095] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0096] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The above-mentioned program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0097] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0098] The above embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

[0099] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent management method for breeding medicinal silkworms, characterized in that: The method comprises the following steps: A plurality of specimen collection points are set in the breeding area, and silkworm pupae inoculated with Beauveria bassiana from the same batch are placed at the specimen collection points, and the sensor device combination acquires the monitoring data sequence, monitoring image set and Beauveria bassiana inoculation parameters corresponding to the specimen collection points in real time based on a preset collection interval; Constructing an image recognition model, wherein the image recognition model recognizes silkworm pupae based on the monitoring image set, generates a classified and annotated image set, outputs a sequence of rigidity of silkworm pupae based on the classified and annotated image set, calculates the rigidity period of silkworm pupae based on the rigidity sequence, and combines the monitoring data sequence, the rigidity sequence and the classified and annotated image set into a sequence set based on the rigidity period; Constructing a monitoring model based on a Bayesian network, inputting all the sequence sets into the monitoring model for analysis based on the collection interval, the sensor device combination and the inoculation parameters, generating a state management model, generating a state estimation table corresponding to the specimen collection point, and constructing a state condition list of silkworm pupae based on all the state estimation tables; Acquire all the collected data corresponding to the current collection time period of the silkworm pupae to be monitored, pre-process the collected data and input them into the state management model, output the rigidification state of the silkworm pupae to be monitored, and judge whether the rigidification state is qualified based on the state condition list. If so, analyze all the collected data corresponding to the next collection time period in real time. Otherwise, output an unqualified condition type, and adjust the control device corresponding to the sensor device combination based on the unqualified condition type. Repeat this step until the silkworm pupae to be monitored complete the rigidification process to form medicinal silkworm pupae.

2. The method according to claim 1, characterized in that The image recognition model identifies silkworm pupae based on the monitoring image set, comprising the following steps: Installing a camera device just above the specimen collection point, the camera device generates monitoring images of silkworm pupae by shooting based on preset shooting parameters and the collection interval, sequentially combining all the monitoring images to generate the monitoring image set, extracting a plurality of the monitoring images from the monitoring image set as initial images, and combining them to generate the initial image set; Converting the initial image into a grayscale image based on the shooting parameters, respectively obtaining the RGB value and pixel coordinates of any pixel in the grayscale image, constructing a three-dimensional coordinate system based on the RGB value, plotting any pixel in the grayscale image in the three-dimensional coordinate system based on the RGB value to generate a spatial point, clustering all the spatial points based on spatial distribution positions using a uniform clustering algorithm to generate multiple cluster clusters, calculating a first distance between any spatial point in the cluster cluster and a cluster center point, and calculating an average value and a standard deviation of all the first distances; Mapping all the spatial points contained in the clusters to the grayscale image corresponding to the initial image based on the pixel coordinates, and performing category marking on the pixels corresponding to all the clusters to generate category labels; The image recognition model is constructed based on a convolutional neural network. The image recognition model is trained on the initial image containing the category label and the remaining monitoring images, and the category labels contained in all the monitoring images are identified based on the recognition scores. The pixels in the monitoring images whose category labels are silkworm pupa labels are identified as silkworm pupa, the monitoring images with the category labels are set as classified annotated images, and all the classified annotated images are combined based on the acquisition interval and set as the classified annotated image set.

3. The method according to claim 2, characterized in that The step of outputting the rigidity degree sequence of silkworm pupae based on the classified and labeled image set comprises the following steps: Obtain pixels identified as silkworm pupae in the classified and labeled image and set them as monitoring pixels, connect adjacent monitoring pixels in sequence based on the pixel coordinates to generate a monitoring area, and set the total number of pixels included in the monitoring area as a first value; Based on the acquisition interval, a center point pixel corresponding to the monitoring area included in the first classified and labeled image in the classified and labeled image set is obtained, and the coordinates of the center point pixel are obtained and set as fixed coordinates, and a second distance between an edge pixel of the monitoring area and the fixed coordinates is obtained based on the pixel coordinates, and an average value corresponding to all the second distances in the monitoring area is calculated and set as a second value; Taking the fixed coordinate as a constant point, calculating the monitoring area corresponding to the constant point in the remaining classified and labeled images in turn to generate the first value and the second value, accumulating all the first values ​​in the classified and labeled images in turn to generate an area value, and setting the average value of all the second values ​​in the classified and labeled images as a growth value; The area values ​​and the growth values ​​corresponding to all the classified and labeled images are counted in sequence based on the acquisition interval to generate a first area sequence and a first growth sequence, and the first area sequence and the first growth sequence are combined and set as the rigidity degree sequence of silkworm pupae.

4. The method according to claim 2, characterized in that: The step of identifying the category labels contained in all the monitoring images based on the recognition scores comprises the following steps: Calculate the recognition score F between any pixel in the grayscale image and the category label based on a first formula, wherein the first formula is: Wherein, D is the first distance of the spatial point corresponding to any pixel in the grayscale image, α is the average value of all the first distances in the cluster, and β is the standard deviation of all the first distances; If the recognition score is less than or equal to the first threshold, it is determined that the classification of the spatial points contained in the cluster is accurate, and any pixel in the grayscale image belongs to the same category label as the cluster; otherwise, the parameters of the uniform clustering algorithm are adjusted and all the spatial points are re-clustered until the recognition score is less than or equal to the first threshold.

5. The method according to claim 1, characterized in that The calculating of the rigidification period of silkworm pupae based on the rigidification degree sequence comprises: The rigidity degree sequence is fitted to generate a first type of curve, an inflection point of the first type of curve is marked, and the first type of curve is divided into a plurality of rigid extension areas and rigid drying areas based on the inflection point, and a time period included in the rigid extension area and the next adjacent rigid drying area is obtained, and the time period is set as the rigidity period.

6. The method according to claim 1, characterized in that The state management model is generated based on the following steps: The monitoring model extracts multiple conditional variables based on the data information contained in the sequence set, performs feature selection on the data information based on the data type to generate a feature quantity set, sets the conditional variables as child nodes, calculates the conditional probability between any of the child nodes and a preset parent node based on the feature quantity set, and performs modeling based on the conditional probability, the preset parent node and the child node to generate the state management model.

7. The method according to claim 6, characterized in that Construct a list of silkworm chrysalis status conditions based on the following steps: Respectively obtaining the category label and the total number of silkworm pupae with the category label in the classified and labeled image and setting them as a third value and a fourth value, setting a plurality of growth state types based on the ratio between the fourth value and the third value, and setting the growth state type as the preset parent node; Acquire the state information of the child node, summarize the first occurrence number of any of the preset parent nodes and the second occurrence number of the same state information corresponding to any of the preset parent nodes, and set the ratio between the second occurrence number and the first occurrence number as the conditional probability between the child node and the preset parent node; Based on the time feature, the preset parent node, the child node and the conditional probability are combined to generate the state estimation table, all the state estimation tables are summarized to generate a first data table, the conditional probabilities corresponding to the same preset parent node and the same child node are averaged and summarized to generate a second data table, and the preset parent node in the normal rigid state is extracted from the second data table and combined into a state condition list of silkworm pupae.

8. The method according to claim 7, characterized in that Judging whether the rigid state is qualified based on the state condition list comprises the following steps: The collected data includes a data sequence corresponding to the sensor device combination, a rigid state corresponding to the collected data is obtained based on the image recognition model and the state management model, and list information corresponding to the rigid state is checked in the second data table based on the collection time period; If the list information belongs to the normal rigid state of the state condition list, the rigid state is determined to be qualified; otherwise, different sub-node items of the list information and the normal rigid state are screened out based on the current collection time period, and the condition variables corresponding to the different sub-node items are set to the unqualified condition type.

Images