Search

CN-121970710-A - Intelligent feeding system and method based on multi-mode space-time fusion and self-evolution correction

CN121970710ACN 121970710 ACN121970710 ACN 121970710ACN-121970710-A

Abstract

The invention relates to the technical field of intelligent aquaculture and provides an intelligent feeding method based on multi-mode space-time fusion and self-evolution correction, which comprises the following steps of system initialization and multi-source parameter setting; the method comprises the steps of receiving and processing multi-mode sensing data, synchronously acquiring the multi-mode sensing data and reading hardware states, carrying out audio-visual signal enhancement processing and feature extraction to accurately extract visual and auditory effective features corresponding to fish swarm ingestion activities, eliminating environment and equipment interference, quantifying the features and generating ingestion indexes, carrying out multi-physical field decoupling and net metabolism oxygen consumption rate inversion, carrying out multi-mode information fusion and safety gating based on variety adaptation weights, and carrying out closed-loop feeding control based on final fusion indexes and variety rhythm strategies.

Inventors

  • WU YIHUA
  • XIAO ZHITONG
  • Liu Gaohao
  • YI XIN
  • ZHANG HANG

Assignees

  • 湖南农业大学

Dates

Publication Date
20260505
Application Date
20260401

Claims (10)

  1. 1. The intelligent feeding method based on multi-mode space-time fusion and self-evolution correction is characterized by comprising the following steps of: Step one, system initialization and multi-source parameter setting Step two, synchronously acquiring multi-mode sensing data and reading hardware states; the third step, audio-visual signal enhancement processing and feature extraction, in order to accurately extract the visual and auditory effective features corresponding to the fish swarm ingestion activity, the environment and equipment interference needs to be eliminated, then the features are quantized and ingestion indexes are generated, and the method is specifically carried out sequentially through the following steps: (3-1) judging the state of the aerator and loading a dynamic interference mask; (3-2), farneback dense optical flow method to calculate the pixel motion vector field; (3-3) calculating the disturbance energy of the water surface; (3-4) normalizing the visual ingestion index; (3-5), sound signal beam forming and aerator noise suppression; (3-6), spectral subtraction processing, and noise suppression; (3-7), calculating and normalizing the auditory feeding index; step four, decoupling of multiple physical fields and inversion of net metabolic oxygen consumption rate; Step five, multi-mode information fusion and safety gating based on variety adaptation weights; Step six, closed-loop feeding control based on the final fusion index and variety rhythm strategy; Step seven, monitoring the feeding process and limiting the total safety; And step eight, self-evolution correction and self-calibration in the whole life cycle.
  2. 2. The intelligent feeding method based on multi-mode space-time fusion and self-evolution correction according to claim 1 is characterized in that firstly, a user inputs geographic information and salinity value of a culture water body through an interactive interface, a system loads corresponding saturated dissolved oxygen calculation standard according to the input geographic information and salinity value, the user selects a current culture variety from a preset culture variety list, and the system loads corresponding feature vectors from a built-in variety characteristic adaptation database according to the selected variety.
  3. 3. The intelligent feeding method based on multi-mode space-time fusion and self-evolution correction according to claim 1, wherein the method is characterized by comprising the following steps of synchronously acquiring multi-source time sequence data, acquiring a video image sequence of a water surface of a feeding area by a visual perception unit, filtering specular reflection light of the water surface by a linear polarization filter integrated in front of a lens during acquisition, acquiring sound signals of the feeding area and the periphery by an auditory perception unit through a linear array formed by a plurality of micro-electro-mechanical systems (MEMS) microphones, and acquiring the dissolved oxygen concentration of a water body by an environment monitoring unit in real time Water temperature And the real-time atmospheric pressure is collected by an on-board atmospheric pressure sensor Executing linkage layer to collect running state signal of aerator in real time And performing time stamp alignment and buffering on all acquired data through an edge computing terminal to form a space-time synchronous multi-mode data stream.
  4. 4. The intelligent feeding method based on multi-mode space-time fusion and self-evolution correction according to claim 1, wherein the third step is to accurately extract visual and auditory effective features corresponding to the feeding activities of the fish shoals, eliminate the environmental and equipment interference, quantify the features and generate feeding indexes; The (3-1) and edge computing terminals read the real-time state of the aerator firstly To judge whether the aerator is started or not, when the aerator is started, in order to eliminate the interference of the water surface disturbance generated by the aerator on the visual signal, the system loads a preset dynamic interference mask in the image coordinate system ; The step (3-2) is that for the pixels of the out-of-mask area, the system calculates the motion vector field of the pixels using Farneback dense optical flow method Namely, the motion speed and the motion direction of each pixel point in the image are measured, and the Farneback optical flow method obtains the displacement vector of each pixel by deducing the motion information of each pixel based on the change of the gray value of the image; The step (3-3) is that the system calculates the disturbance energy of the water surface based on the pixel motion vector field calculated by Farneback optical flow method The specific calculation formula is as follows: Wherein, the Represents the disturbance energy of the water surface, For masking in pixels The value at which the value is to be calculated, Is the square of the motion vector field of the pixel, representing the disturbance energy at that location, When the value of (2) is zero, the position is in the mask, the movement energy is not calculated by disturbance energy, if If the position is one, the position is an effective area, and disturbance energy is counted; The (3-4) is calculated to obtain the water surface disturbance energy Normalizing to obtain visual ingestion index The normalization process is accomplished by the following formula: normalized visual ingestion index Between 0 and 1, a larger value indicates more significant disturbance on the water surface, indicating that the fish school feeding activity is more active; the system adopts a delay summation beam forming algorithm to process the sound signals received by the microphone array (3-5). Delay-and-sum beamforming algorithms adjust the phase delay of each microphone channel The signals received by a plurality of microphones are weighted and combined to form a beam pointing to a specific direction, the beam forming aims at concentrating the beam in the direction of a feeding drop point, and the signals after beam forming Calculated by the following formula: Wherein, the Is a signal after the beam forming, Is the first The signals received by the microphones are then transmitted to the respective microphones, Is the corresponding weighting coefficient of the weight of the sample, Is the first The phase of the individual microphones is delayed and, Time is; the (3-6) beamformed signals Further performing spectrum subtraction treatment to remove steady-state mechanical noise generated by the aerator and other equipment; The (3-7) processing the acoustic signals, the system forming the beamformed signals within the acoustic characteristic frequency band specified by the variety characteristic vector Integrating, and calculating the energy of biological water impact sound, wherein the specific integrating operation is as follows: Wherein, the Is the energy of the signal, the integral operation is to integrate the signal in a specified frequency band, calculate the total energy of the biological water hammer sound, and the auditory ingestion index after normalization Also between 0 and 1, a larger value indicates a stronger activity of the aquatic organism, indicating that the fish population is more active in feeding.
  5. 5. The intelligent feeding method based on multi-mode space-time fusion and self-evolution correction according to claim 1, wherein the fourth step is that the edge computing terminal calls a multi-physical-field decoupling engine to compute saturated dissolved oxygen values corrected by air pressure, temperature and salinity : Wherein, the Is the air pressure of the air, and the air pressure is the air pressure, Is the saturated water vapor pressure of water, which is related to the temperature In relation to each other, Is a temperature correction function that represents the effect of temperature on dissolved oxygen; Is the salinity value of the water body, Is a salinity correction constant for correcting the effect of salinity on dissolved oxygen; based on aerator state Oxygenation capacity of aerator And the basic oxygen consumption rate of the water body Inversion of net metabolic oxygen consumption rate of fish shoal The calculation formula is as follows: Wherein, the The time change rate of the concentration of the dissolved oxygen is represented and reflects the oxygen consumption rate in the water body; Is the dissolved oxygen concentration measured in real time, The corrected saturated dissolved oxygen concentration is compared with the influence factors of oxygen consumption by the measured dissolved oxygen variation to reversely deduce the net metabolic oxygen consumption rate of the fish school The oxygen consumption of the fish shoal in unit time can reflect the physiological activities and ingestion demands of the fish shoal, and a basis is provided for intelligent feeding.
  6. 6. The intelligent feeding method based on multi-modal space-time fusion and self-evolution correction according to claim 1, wherein the fourth step is to calculate After normalization, chemical feeding index is obtained For quantifying the feeding behaviour of a fish population, the normalization process is accomplished by the following formula: Wherein, the Is the maximum of all historic moments A value that ensures a chemical feeding index between 0 and 1, a larger value indicating more active feeding behavior of the fish population, A feeding demand index for a fish school based on metabolism and oxygen consumption is provided.
  7. 7. The intelligent feeding method based on multi-modal space-time fusion and self-evolution correction according to claim 1, wherein the fifth step is that the edge computing terminal corrects the coefficient according to the weight in the variety feature vector loaded in the first step 、 、 Visual ingestion index Auditory ingestion index Chemical feeding index Weighting and fusing to generate primary fused feeding index : Calling a fault safety management module to evaluate the safety state and monitor the atmospheric pressure Checking the continuity and validity of each sensor data; if the primary weather risk or the secondary sensor fault is judged to exist, a degradation safety gating coefficient is generated ; Will initially fuse the feeding index And safety gating coefficient Multiplying to obtain the final fusion index for feeding decision I.e. 。
  8. 8. The intelligent feeding method based on multi-modal space-time fusion and self-evolution correction according to claim 7, wherein the step six is that the system judges the final fusion index Whether or not the starting threshold set by the variety is exceeded If (1) Controlling the batch feeder to keep in a stop state if According to the proportional gain coefficient Calculating a real-time duty ratio control signal of a feeding motor ; Wherein the method comprises the steps of For minimum start duty cycle, calculate the result Is limited to 0% -100% and calculated And (3) modulating the control signal according to the feeding rhythm control mode corresponding to the variety loaded in the step one.
  9. 9. The intelligent feeding method based on multi-mode space-time fusion and self-evolution correction according to claim 7, wherein in the step seven, the system calibrates the flow rate according to the feeding machine in the feeding execution process And a real-time duty cycle The accumulated feeding amount is calculated by real-time integration, and the integral formula is as follows: Wherein, the Refers to the starting time of the feeding at this time, The accumulated feeding amount is compared with the single maximum feeding amount upper limit calculated according to the estimated weight and the water temperature of the pond fish at the current moment Comparing if the accumulated feeding amount reaches Namely satisfy When, then, no matter the final fusion index The system forcibly stops feeding this time when the height is high or low; And step eight, in the set non-feeding period, the system automatically starts a self-calibration process to update model data, firstly, the aerator is closed, and basic data which can reflect the natural oxygen consumption rate of the water body under the condition of no fish shoal is updated by analyzing and fitting the change trend of the content of dissolved oxygen in the water body along with the time.
  10. 10. The intelligent feeding system based on the multi-mode space-time fusion and the self-evolution correction is characterized by comprising a system initialization and parameter configuration module, a multi-mode data acquisition module, an audiovisual signal processing module, a multi-physical field decoupling module, a multi-mode fusion module, a closed-loop feeding control module, a safety monitoring module, a self-evolution correction module and a non-feeding period updating core parameters, wherein the system initialization and parameter configuration module loads regional saturated dissolved oxygen calculation reference and variety characteristic vectors, the multi-mode data acquisition module synchronously acquires vision, hearing, environment and equipment state data, the audiovisual signal processing module extracts vision feeding indexes and hearing feeding indexes, the multi-physical field decoupling module inverts net metabolic oxygen consumption rate of fish shoals and generates chemical feeding indexes, the multi-mode fusion module is combined with safety gating to obtain a final fusion index, the closed-loop feeding control module performs feeding according to a feeding rhythm control mode, and the safety monitoring module monitors accumulated feeding quantity and equipment state.

Description

Intelligent feeding system and method based on multi-mode space-time fusion and self-evolution correction Technical Field The invention relates to the technical field of intelligent aquaculture, in particular to an intelligent feeding system and method based on multi-mode space-time fusion and self-evolution correction. Background The aquaculture feeding is a core technical operation in an aquaculture production process, and refers to production behaviors that aquaculture personnel put in adaptive feed to aquaculture water at fixed time, fixed quantity and fixed point according to biological characteristics, growth stages, nutritional requirements and environmental parameters of aquaculture water of aquaculture objects, so as to meet ingestion requirements of the aquaculture objects and promote healthy and rapid growth of the aquaculture objects. At present, the traditional aquaculture feeding technology mainly comprises three implementation modes, namely manual feeding, wherein a farmer judges the feeding state of the shoal of fish, the feeding amount of the baits and the feeding rhythm by means of daily experience, the unified quantitative standard is not needed, the method completely depends on personal subjective judgment, and the mechanical quantitative feeding is realized, the feeding machine is controlled by a timer or a simple switch, the baits are fed according to fixed time and fixed dose, and the real-time feeding requirement and the environmental change of the shoal of fish are not considered. The traditional aquaculture feeding mode has the obvious defects that manual feeding completely depends on subjective experience of aquaculture personnel, uniform fish swarm feeding state judgment, bait feeding amount and feeding rhythm quantification standard are not available, overfeeding or shortness is easy to occur due to experience difference, efficiency is low, large-scale aquaculture cannot be adapted, mechanical quantitative feeding realizes fixed-time and fixed-dose bait feeding only through a timer or a simple switch, real-time feeding requirements of fish swarms and dynamic change of aquaculture environment are lacked, flexibility is extremely poor, bait waste and water pollution are easy to be caused, fish growth is influenced due to the fact that feeding is not performed in time, and basic requirements of accurate aquaculture are difficult to meet. Therefore, we propose an intelligent feeding system and method based on multi-modal space-time fusion and self-evolution correction. Disclosure of Invention The invention provides an intelligent feeding system and method based on multi-mode space-time fusion and self-evolution correction, which solve the problems that in the related technology, manual feeding depends on subjective experience, has no unified quantization standard, is low in efficiency and cannot adapt to large-scale cultivation, fixed time and dose feeding is performed by mechanical quantitative feeding, consideration of real-time feeding requirements and environment change of fish shoals is lacked, flexibility is poor, bait waste and water pollution are easy to be caused, basic intelligent feeding depends on single perception dimension only, has no region and variety adaptation mechanism, has no self-updating capability of core parameters, cannot dynamically adapt to cultivation environment and fish shoals habit change, and further is difficult to realize accurate feeding, cultivation cost is saved and water environment is protected. The technical scheme of the invention is as follows: An intelligent feeding method based on multi-mode space-time fusion and self-evolution correction comprises the following steps: Step one, system initialization and multi-source parameter setting Step two, synchronously acquiring multi-mode sensing data and reading hardware states; the third step, audio-visual signal enhancement processing and feature extraction, in order to accurately extract the visual and auditory effective features corresponding to the fish swarm ingestion activity, the environment and equipment interference needs to be eliminated, then the features are quantized and ingestion indexes are generated, and the method is specifically carried out sequentially through the following steps: (3-1) judging the state of the aerator and loading a dynamic interference mask; (3-2), farneback dense optical flow method to calculate the pixel motion vector field; (3-3) calculating the disturbance energy of the water surface; (3-4) normalizing the visual ingestion index; (3-5), sound signal beam forming and aerator noise suppression; (3-6), spectral subtraction processing, and noise suppression; (3-7), calculating and normalizing the auditory feeding index; step four, decoupling of multiple physical fields and inversion of net metabolic oxygen consumption rate; Step five, multi-mode information fusion and safety gating based on variety adaptation weights; Step six, closed-loop feeding control base