Deep Learning-Based Bullfrog Dynamic Disease Recognition System and Method

Abstract

The present invention relates to the technical field of bullfrog breeding, and specifically discloses a dynamic disease recognition system and method for bullfrogs based on deep learning, including setting a monitoring area and arranging a number of driving devices in the monitoring area to drive the bullfrogs in the monitoring area within a preset time period; obtaining first video information in the monitoring area within a preset time period after the driving device is started and second video information during the feeding time period, and analyzing the first video information and the second video information to carry out disease early warning based on the jumping information and moving information. The present invention judges the diseases that can be reflected in the appearance among the common diseases of bullfrogs through deep learning and recognition of pictures. For the diseases that cannot be reflected in the appearance, through the jumping information and moving information of bullfrogs during the driving and feeding processes, and based on the jumping information and moving information, disease early warning is carried out to achieve a better bullfrog breeding management effect.

Description

Technical Field

[0001] The present invention relates to the technical field of bullfrog farming, and specifically, to a dynamic disease recognition system and method for bullfrogs based on deep learning. Background Art

[0002] Bullfrog farming has a long history, and disease prevention and control in the farming process has always been one of the research directions in the farming industry. At present, the method of identifying diseases by recognizing the appearance of diseased bullfrogs through a large model trained by deep learning has been applied to a certain extent.

[0003] The problem is that, on the one hand, the diseases of some bullfrogs do not manifest in their appearance, so it is difficult to use the large model obtained by deep learning. On the other hand, the results obtained by image recognition often occur after the fact, and for bullfrog farming, it can only achieve damage control in a timely manner and is difficult to give early warnings.

[0004] In view of this, the purpose of the present invention is to provide a dynamic disease recognition system and method for bullfrogs based on deep learning, which can give a certain degree of early warning and recognition of bullfrog diseases based on the dynamic state of bullfrogs, and improve the management efficiency of bullfrog farming. Summary of the Invention

[0005] The purpose of the present invention is to provide a dynamic disease recognition system and method for bullfrogs based on deep learning, and solve the following technical problems:

[0006] How to give a certain degree of early warning and recognition of bullfrog diseases based on the dynamic state of bullfrogs.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] A dynamic disease recognition method for bullfrogs based on deep learning includes:

[0009] Set a monitoring area and set a number of driving devices in the monitoring area to drive the bullfrogs in the monitoring area within a preset time period.

[0010] Obtain the first video information in the monitoring area within a preset time period after the driving device is started and the second video information during the feeding time period, and analyze the first video information and the second video information to obtain the first effective image and the second effective image.

[0011] Perform disease recognition based on the first effective image.

[0012] Obtain the jumping information and moving information through the second effective image and perform disease early warning based on the jumping information and the moving information.

[0013] Through the above technical solutions, a method for disease identification and early warning based on image information during the driving and feeding of bullfrogs is provided. Specifically, for diseases that can be manifested in the appearance among common bullfrog diseases, judgment is made through deep learning and recognition of pictures. For diseases that cannot be manifested in the appearance, disease early warning is carried out through the jumping information and moving information of bullfrogs during the driving and feeding processes, and a better bullfrog breeding management effect is achieved.

[0014] As a further technical solution of the present invention, the steps of obtaining the first effective image and performing disease identification based on the first effective image include:

[0015] Identify and mark complete bullfrog feature images from all video frames of the first video information and the second video information, and record the number of marks;

[0016] Select several video frames with the largest number of the same marks as the first effective image;

[0017] Set up a trained bullfrog disease identification model, input the first effective image into the corresponding bullfrog disease identification model, and perform disease identification by the bullfrog disease identification model.

[0018] As a further technical solution of the present invention, the steps of obtaining the second effective image and obtaining jumping information through the second effective image include:

[0019] Obtain video frames containing bullfrog jumping actions from all video frames of the first video information and the second video information;

[0020] Take at least three video frames including one bullfrog jumping starting point, jumping ending point, and jumping highest point as the second effective image;

[0021] Construct a jumping curve of the current bullfrog based on the second effective image, and record the number of jumping curves.

[0022] As a further technical solution of the present invention, the steps of obtaining moving information through the second effective image include:

[0023] Select several groups of reference images from the second effective image in the same scene;

[0024] Partition the frog group along the moving direction of the frog group in the reference image, and after partitioning merge the partition with the adjacent partition to obtain several target partitions;

[0025] Obtain the average moving speed of the frog group in the target partition based on the time interval and moving distance of the frog group in the target partition within a group of reference images.

[0026] As a further technical solution of the present invention, the step process of disease early warning based on jump information and movement information is as follows:

[0027] Through the formula:

[0028]

[0029] Obtain the activity coefficient Ac of the current day, where α 1 is a preset first proportion coefficient, α 2 is a preset second proportion coefficient, Ju 1 is the first jump coefficient, Ju 2 is the second jump coefficient, n 1 is the number of non-repeated jump curves occurring in all the first video information and second video information obtained on the current day, T is the breeding time, n 0 is a preset jump curve number selection function based on the breeding time, g is a normalization function, β 1 is a preset first jump weight coefficient, β 2 is a preset second jump weight coefficient, γ 1 is a preset first movement weight coefficient, γ 2 is a preset second movement weight coefficient, Mo 1 is the first movement coefficient, Mo 2 is the second movement coefficient, σ is a correction amount selection function;

[0030] Continuously count the activity coefficient and obtain the judgment coefficient based on historical data, and early warn of bullfrog diseases based on the judgment coefficient.

[0031] Through the above technical solution, the step process of disease early warning based on jump information and movement information is provided. Specifically, the present invention obtains the first jump coefficient and the second jump coefficient related to jumps through jump curves and the number of jumps, and counts the jumping situations of bullfrogs in the feeding and driving states. For some diseases of bullfrogs that do not manifest in appearance, they will affect the appetite or motor ability of bullfrogs, and most of the diseases in the breeding process are infectious diseases. Through the identification and estimation of the overall motor ability, the disease state of bullfrogs can be early warned, so as to avoid breeding losses caused by the large-scale spread of diseases that do not manifest in appearance.

[0032] As a further technical solution of the present invention, the obtaining processes of the first jump coefficient, the second jump coefficient, the first movement coefficient, and the second movement coefficient are as follows:

[0033] Through the formula:

[0034]

[0035] Obtain the first jump coefficient Ju 1, the second jump coefficient Ju 2 and the first movement coefficient Mo 1 , the second movement coefficient Mo 2 , where f i is the function expression of the i-th jump curve within a preset time period, t i is the time corresponding to the starting point of the i-th jump curve, is the time corresponding to the ending point of the i-th jump curve, is the preset first jump curve, f j is the function expression of the j-th jump curve within the feeding time period, and are respectively the time starting point and time ending point set for the preset jump curve, t j is the time corresponding to the starting point of the j-th jump curve, is the time corresponding to the ending point of the j-th jump curve, is the preset second jump curve, and are respectively the time starting point and time ending point set for the preset jump curve, m 1 is the total number of jump curves within the preset time period, m 2 is the total number of jump curves during the feeding time period, i is a natural number with a value range between 1 and m 1 and j is a natural number with a value range between 1 and m 2 .

[0036] As a further technical solution of the present invention: The process of obtaining the judgment coefficient includes:

[0037] Through the formula:

[0038]

[0039] Obtain the judgment coefficient Jg of the current day, where Ac is the activity coefficient of the current day, Ac k is the activity coefficient of the k-th comparison day in the historical data, N is the total number of selected comparison days, and τ is a preset compensation function used to obtain the compensation value according to the breeding time T.

[0040] As a further technical solution of the present invention: The process of warning about bullfrog diseases based on the judgment coefficient includes:

[0041] Compare the judgment coefficient with the preset judgment interval [Jg 1 , Jg 2 . If Jg > Jg 2 , then it is determined that the frog group is abnormal and early warning is required;

[0042] If Jg < Jg 2 , then it is determined that the frog group is normal and early warning is not required;

[0043] If Jg falls within the judgment interval, obtain the second jump coefficient of the feeding time period, compare the second jump coefficient with a plurality of preset safety intervals, and if the second jump coefficient falls within the safety interval, no warning is required; otherwise, an alarm is issued.

[0044] Through the above technical solutions, a solution for obtaining a judgment coefficient and giving an alarm based on the judgment coefficient is provided. Specifically, the judgment coefficient of the present invention is obtained based on the daily activity coefficient and historical data, and can reflect the change of the daily activity coefficient compared with the reference day, so as to give an alarm. In addition, a judgment interval is set. When the judgment coefficient falls within the judgment interval, it means that the change of the daily activity coefficient compared with the reference day is not sufficient to judge whether an alarm is needed. At this time, through the second jump coefficient obtained during the daily feeding time period, it can be judged whether the data of the bullfrogs in the feeding state is normal for secondary verification, thereby improving the comprehensiveness of the alarm.

[0045] The present invention also provides a dynamic disease recognition system for bullfrogs based on deep learning, including:

[0046] A monitoring module for video monitoring of the monitoring area;

[0047] A driving module for driving the frog group;

[0048] A judgment module for judging bullfrog diseases and whether an alarm is needed;

[0049] An alarm module for remotely notifying the management personnel when it is judged that an alarm is needed. The beneficial effects of the present invention:

[0050] (1) For diseases of bullfrogs that can be reflected in the appearance in common diseases of bullfrogs, judgment is made through deep learning and recognition of pictures. For diseases that cannot be reflected in the appearance, through the jumping information and movement information of bullfrogs during driving and feeding, and disease warning is carried out based on the jumping information and movement information, so as to achieve a better bullfrog breeding management effect.

[0051] (2) The present invention obtains the first jump coefficient and the second jump coefficient related to jumping through the jump curve and the number of jumps, and counts the jumping conditions of bullfrogs in the feeding and driving states. In some diseases of bullfrogs that are not reflected in the appearance, the appetite or motor ability of bullfrogs will be affected, and most of the diseases in the breeding process are infectious diseases. Through the recognition and estimation of the overall motor ability, the disease state of bullfrogs can be warned, thereby avoiding breeding losses caused by the large-scale spread of diseases that are not reflected in the appearance.

[0052] (3) The judgment coefficient of the present invention is obtained based on the daily activity coefficient and historical data, which can reflect the change of the daily activity coefficient compared with the control day, so as to give an early warning. In addition, a judgment interval is set. When the judgment coefficient falls within the judgment interval, it indicates that the change of the daily activity coefficient compared with the control day is not sufficient to judge whether an early warning is needed. At this time, the second jump coefficient obtained through the daily feeding time period is used to judge whether the data of the bullfrogs in the feeding state is normal for secondary verification, so as to improve the comprehensiveness of the early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The present invention will be further described below with reference to the accompanying drawings.

[0054] Figure 1 is the overall step flow chart of the present invention;

[0055] Figure 2 is the flow chart of the first effective image acquisition step of the present invention;

[0056] Figure 3 is the flow chart of the second effective image acquisition step of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0057] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0058] Please refer to Figures 1 - 3 As shown, in one embodiment, a method for identifying dynamic diseases of bullfrogs based on deep learning is provided, including:

[0059] S100. Set a monitoring area and set a number of driving devices in the monitoring area to drive the bullfrogs in the monitoring area during a preset time period. It should be noted that the bullfrogs in the same monitoring area need to be in the same growth cycle to maintain a similar growth state;

[0060] S200. Obtain the first video information in the monitoring area during the preset time period after the driving device is started and the second video information during the feeding time period, and analyze the first video information and the second video information to obtain the first effective image and the second effective image;

[0061] S300. Identify diseases according to the first effective image;

[0062] S400. Obtain jump information and movement information through the second effective image and perform disease early warning based on the jump information and the movement information.

[0063] In this embodiment, a method for disease identification and early warning based on image information during the driving and feeding of bullfrogs is provided. Specifically, for diseases of bullfrogs that can be reflected in appearance among common diseases, judgment is made through deep learning and recognition of pictures. For diseases that cannot be reflected in appearance, disease early warning is carried out based on the jumping information and movement information of bullfrogs during the driving and feeding processes, so as to achieve better bullfrog breeding management effects.

[0064] The steps of obtaining the first effective image and performing disease identification based on the first effective image include:

[0065] S310. Identify and mark complete bullfrog feature images from all video frames of the first video information and the second video information, and record the marked quantity. The feature images include head images, abdominal images, and back images. It should be noted that in order to reduce computing power waste, an attention mechanism can be set, and key attention targets are identified before identifying and marking complete bullfrog feature images;

[0066] S320. Select several video frames with the largest number of the same marks as the first effective images;

[0067] S330. Set a trained bullfrog disease identification model, which is obtained through deep learning training with case pictures of common diseases such as hemorrhagic disease, tilted head disease, skin ulcer disease, bloated belly disease, flatulence disease, and erythema disease that can cause external physical sign changes in bullfrogs. Input the first effective image into the corresponding bullfrog disease identification model, and the bullfrog disease identification model performs disease identification and outputs the identification result. It should be noted that the minimum number of identifications can be set. If the number of bullfrogs with diseases identified in the first effective image is not higher than the minimum number, no early warning is carried out, so as to avoid false warning problems caused by the abnormalities of individual bullfrogs.

[0068] The steps of obtaining the second effective image and obtaining jumping information through the second effective image include:

[0069] S340. Obtain video frames containing bullfrog jumping actions from all video frames of the first video information and the second video information;

[0070] S350. Use at least three video frames including one bullfrog jumping starting point, jumping ending point, and jumping highest point as the second effective images;

[0071] S360. Construct the jumping curve of the current bullfrog based on the second valid image and record the number of jumping curves. The jumping information includes the jumping curve and the number of jumping curves. It should be noted that, in order not to record the jumping curve repeatedly, an identification vector can be constructed based on the appearance of the bullfrog. After obtaining the jumping curve, the current curve can be matched with the identification vector. In this way, when the same bullfrog within the same jumping curve is recognized in subsequent pictures, it can be skipped to avoid repeated recording.

[0072] The steps of obtaining the movement information through the second valid image include:

[0073] S410. In the same scenario, select several groups of reference images from the second valid image. The same scenario means that the valid images are selected from the same preset time period or feeding time period. In addition, the reference pictures carry timestamps. A group of reference pictures consists of two pictures. The time interval between the two pictures can be obtained through the timestamp. Obviously, the shooting angles and positions of a group of reference photos are preferably the same. If they are different, coordinate conversion needs to be performed first.

[0074] S420. In the reference images, partition the frog group along the movement direction of the frog group, and merge the partitioned partition with the adjacent partition to obtain several target partitions. The value range of a is between 20 and 50. In this embodiment, it is set to 30. The smaller the value of a, the more bullfrogs can be included in one target partition, and at the same time, the fewer the number of target partitions. It is necessary to ensure that the number of target partitions is sufficient to provide enough data for analysis.

[0075] S430. Based on the time interval and movement distance of the frog group in a group of reference images within the target partition, obtain the average moving speed of the frog group in the target partition.

[0076] The process of disease early warning based on the jumping information and movement information is as follows:

[0077] Through the formula:

[0078]

[0079] Obtain the activity coefficient Ac of the current day, where α 1 is the preset first specific gravity coefficient, α 2 is the preset second specific gravity coefficient, Ju 1 is the first jumping coefficient, Ju 2 is the second jumping coefficient, n 1 is the number of non-repeated jumping curves that occur in all the first video information and second video information obtained on the current day, T is the breeding time, n 0 is the preset jumping curve number selection function set based on the breeding time, g is the normalization function, and its value range is between 0 and 1, β 1is the preset first jump weight coefficient, β 2 is the preset second jump weight coefficient, γ 1 is the preset first movement weight coefficient, γ 2 is the preset second movement weight coefficient, Mo 1 is the first movement coefficient, Mo 2 is the second movement coefficient, σ is the correction amount selection function, preferably a look-up table function, where α 1 、α 2 、β 1 、β 2 and γ 1 、γ 2 are all constants set based on empirical data;

[0080] Continuously count the activity coefficient and obtain the judgment coefficient based on historical data, and give an early warning of bullfrog diseases based on the judgment coefficient.

[0081] In this embodiment, a step process for disease early warning based on jump information and movement information is provided. Specifically, the present invention obtains the first jump coefficient and the second jump coefficient related to jumps through the jump curve and the number of jumps, and counts the jumping situations of bullfrogs in the feeding and driving states. Diseases in bullfrogs that are not reflected in the appearance will affect the appetite or motor ability of bullfrogs, and most of the diseases in the breeding process are infectious diseases. Through the identification and estimation of the overall motor ability, the disease state of bullfrogs can be warned, thus avoiding breeding losses caused by the large-scale spread of diseases that are not reflected in the appearance.

[0082] Among them, the acquisition processes of the first jump coefficient, the second jump coefficient, the first movement coefficient, and the second movement coefficient are as follows:

[0083] Through the formula:

[0084]

[0085] Obtain the first jump coefficient Ju 1 、the second jump coefficient Ju 2 and the first movement coefficient Mo 1 、the second movement coefficient Mo 2 , where f i is the function expression of the i-th jump curve within a preset time period. It should be noted that the function expression is constructed with time as the x-axis and distance as the y-axis, and the starting point of the jump curve, that is, the position where the takeoff action is located, is set to zero, t i is the corresponding time of the starting point of the i-th jump curve, is the corresponding time of the end point of the i-th jump curve, is the preset first jump curve, f jis the function expression of the j-th jump curve within the feeding time period, and are the time start point and time end point set for the preset jump curve respectively, and t j is the time corresponding to the start point of the j-th jump curve, is the time corresponding to the end point of the j-th jump curve, is the preset second jump curve, and are the time start point and time end point set for the preset jump curve respectively, and m 1 is the total number of jump curves within the preset time period, and m 2 is the total number of jump curves in the feeding time period. i is a natural number with a value range between 1 and m 1 and j is a natural number with a value range between 1 and m 2

[0086] The process of obtaining the judgment coefficient includes:

[0087] Through the formula:

[0088]

[0089] Obtain the judgment coefficient Jg of the current day, where Ac is the activity coefficient of the current day, and Ac k is the activity coefficient of the k-th comparison day in the historical data. N is the total number of selected comparison days. τ is a preset compensation function, set based on the historical data, used to obtain the compensation value according to the breeding time T. The comparison day data is the past breeding data in the selected historical data, usually set as the previous N days. If there are abnormal data in the previous N days, it can be appropriately increased or maintained.

[0090] The process of warning about bullfrog diseases based on the judgment coefficient includes:

[0091] Compare the judgment coefficient with the preset judgment interval [Jg 1 , Jg 2 . If Jg > Jg 2 then judge that the frog group is abnormal and a warning is needed;

[0092] If Jg < Jg 2 , then judge that the frog group is normal and no warning is needed;

[0093] If Jg falls within the judgment interval, obtain the second jump coefficient in the feeding time period, compare the second jump coefficient with multiple preset safety intervals. If the second jump coefficient falls within the safety interval, no warning is needed, otherwise an alarm is sent to the management personnel.

[0094] ​In this embodiment, a solution for obtaining a judgment coefficient and giving an early warning based on the judgment coefficient is provided. Specifically, the judgment coefficient of the present invention is obtained based on the daily activity coefficient and historical data, which can reflect the change of the daily activity coefficient compared with the control day, so as to give an early warning. In addition, a judgment interval is set. When the judgment coefficient falls within the judgment interval, it indicates that the change of the daily activity coefficient compared with the control day is not sufficient to judge whether an early warning is needed. At this time, the second jump coefficient obtained during the daily feeding period is used to judge whether the data of the bullfrogs in the feeding state is normal for secondary verification, so as to improve the comprehensiveness of the early warning.

[0095] This embodiment also provides a bullfrog dynamic disease identification system based on deep learning, including:

[0096] A monitoring module for video monitoring of the monitoring area;

[0097] A driving module for driving the frog group. The specific implementation manner of the driving device is not limited. In this embodiment, an automated controlled enclosure hitting module is used as the driving device, and the advantage is that it combines vibration and sound and is easy to control;

[0098] A judgment module for judging bullfrog diseases and whether an early warning is needed;

[0099] An early warning module for remotely notifying the management personnel when an early warning is needed.

[0100] The above has described a specific embodiment of the present invention in detail, but the described content is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the application of the present invention shall still fall within the scope covered by the patent of the present invention.

Claims

1. A method for dynamic disease identification of bullfrog based on deep learning, characterized in that: The steps include: Setting up a monitoring area and installing a number of driving devices in the monitoring area to drive away bullfrogs in the monitoring area during a preset time period; Acquire first video information in the monitoring area within a preset time period after the driving device is activated and second video information within a feeding time period, and analyze the first video information and the second video information to acquire a first valid image and a second valid image; The steps of acquiring the second effective image and acquiring the jump information through the second effective image include: Acquire a video frame containing a bullfrog jumping action from all video frames of the first video information and the second video information; taking three video frames including at least a bullfrog jumping starting point, a jumping end point and a jumping highest point as the second valid images; Constructing a jumping curve of the current bullfrog based on the second valid image, and recording the number of jumping curves; performing disease identification according to the first valid image; Acquire jump information and movement information through the second effective image and perform disease warning based on the jump information and movement information; The step of acquiring movement information through the second effective image comprises: In the same scene, selecting a plurality of groups of reference images from the second valid image; In the reference image, the frog group is partitioned along the direction of its movement, and after the partition, The partition of is merged with the adjacent partitions to obtain several target partitions, and the adjacent partitions satisfy, The value range of a is between 20-50; The average moving speed of the frog group in the target partition is obtained based on the time interval and moving distance of the frog group in the target partition in a set of reference images; The process of disease early warning based on jump information and movement information is as follows: By formula: Obtain the activity coefficient Ac of the day, where α1 is the preset first weight coefficient, α2 is the preset second weight coefficient, Ju1 is the first jump coefficient, Ju2 is the second jump coefficient, n1 is the number of non-repeated jump curves occurring in all the first video information and the second video information obtained on the day, T is the breeding time, n0 is the preset jump curve number selection function set based on the breeding time, g is the normalization function, β1 is the preset first jump weight coefficient, β2 is the preset second jump weight coefficient, γ1 is the preset first movement weight coefficient, γ2 is the preset second movement weight coefficient, Mo1 is the first movement coefficient, Mo2 is the second movement coefficient, and σ is the correction amount selection function; The activity coefficient is continuously counted and the judgment coefficient is obtained based on historical data, and early warning of bullfrog diseases is issued based on the judgment coefficient.

2. The method for dynamic disease identification of bullfrog based on deep learning according to claim 1, characterized in that: The steps of acquiring a first valid image and identifying a disease according to the first valid image include: Identify and mark complete bullfrog feature images from all video frames of the first video information and the second video information, and record the number of marks; Selecting a number of video frames with the largest number of identical marks as first valid images; A trained bullfrog disease recognition model is set, and the first valid image is input into the corresponding bullfrog disease recognition model, and the bullfrog disease recognition model performs disease recognition.

3. The method for dynamic disease identification of bullfrog based on deep learning according to claim 1, characterized in that: The process of obtaining the first jump coefficient, the second jump coefficient, the first movement coefficient, and the second movement coefficient is as follows: By formula: Obtain the first jump coefficient Ju1, the second jump coefficient Ju2 and the first movement coefficient Mo1, the second movement coefficient Mo2, where f i is the function expression of the i-th jump curve in the preset time period, t i is the starting point corresponding to the time of the i-th jump curve, is the time corresponding to the end point of the i-th jump curve, is the preset first jump curve, f j is the functional expression of the jth jump curve in the feeding period, and Set the start and end time of the preset jump curve respectively, t j is the starting point corresponding to the time of the jth jump curve, is the time corresponding to the end point of the j-th jump curve, is the preset second jump curve, and They are respectively used to set the time start point and time end point for the preset jump curve, m1 is the total number of jump curves in the preset time period, m2 is the total number of jump curves in the feeding time period, i is a natural number ranging from 1 to m1, and j is a natural number ranging from 1 to m2.

4. The method for dynamic disease identification of bullfrog based on deep learning according to claim 3 is characterized in that: The process of obtaining the judgment coefficient includes: By formula: Get the judgment coefficient Jg of the day, where Ac is the activity coefficient of the day, Ac k is the activity coefficient of the kth control day in the historical data, N is the total number of selected control days, and τ is the preset compensation function, which is used to obtain the compensation value according to the breeding time T.

5. The method for dynamic disease identification of bullfrog based on deep learning according to claim 4 is characterized in that: The process of early warning of bullfrog diseases based on the judgment coefficient includes: Compare the judgment coefficient with the preset judgment interval [Jg1, Jg2]. If Jg>Jg2, it is judged that the frog group is abnormal and an early warning is required; If Jg<Jg2, the frog population is judged to be normal and no warning is needed; If Jg falls within the judgment interval, the second jump coefficient of the feeding time period is obtained, and the second jump coefficient is compared with multiple preset safety intervals. If the second jump coefficient falls within the safety interval, no warning is required, otherwise an alarm is issued.

6. A bullfrog dynamic disease identification system based on deep learning, using a bullfrog dynamic disease identification method based on deep learning as claimed in any one of claims 1 to 5, characterized in that: include: Monitoring module, used for video monitoring of the monitoring area; The driving module is used to drive away the frogs; The judgment module is used to judge bullfrog diseases and whether an early warning is needed; The early warning module remotely notifies management personnel when an early warning is deemed necessary.

Images