Duck shed intelligent monitoring method and system based on image recognition

Abstract

The invention discloses an intelligent duck shed monitoring method and system based on image recognition, and the method comprises the steps: obtaining environment monitoring data from an environment monitoring device, obtaining original image data of a breeding region from an image collection device, and processing the original image data through an image enhancement algorithm, thereby obtaining a preprocessed image data set; according to weight parameters output by the duck group behavior and environment correlation model, performing weighted fusion on the environment monitoring data and the preprocessed image data set to generate a multi-modal feature data set; if the group behavior index in the multi-modal feature data set deviates from a preset threshold range, extracting a spatial-temporal feature map through a convolutional neural network, and generating individual behavior trajectory data in combination with a multi-target tracking algorithm; and inputting the individual behavior trajectory data into the lightweight neural network model, and outputting a health state classification result. According to the intelligent duck breeding system, accurate monitoring and intelligent adjustment of the duck breeding environment are achieved, and the breeding efficiency and the duck group health level are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular discloses an intelligent monitoring method and system for a duck house based on image recognition. Background Art

[0002] Poultry farming is an important pillar of agricultural modernization. The application of intelligent monitoring technology is of decisive significance for improving farming efficiency, reducing mortality and optimizing operating costs.

[0003] With the increasing global concern for food safety and animal welfare, intelligent monitoring of duck houses has become a key area in promoting green farming. However, existing methods have obvious limitations in practical applications.

[0004] Many solutions rely on a single data source, such as environmental sensors or simple cameras, which struggle to fully capture the complex interplay between duck flock behavior, health, and environmental changes. Furthermore, traditional monitoring systems often require extensive manual intervention, resulting in low processing efficiency and difficulty adapting to the dynamic demands of large-scale farming. Against this backdrop, intelligent duck house monitoring faces multiple technical challenges.

[0005] The primary challenge is the collection and fusion of multimodal data. Due to the complex environment of the duck house, uneven lighting, and densely packed ducks, image data quality is unstable. A single sensor cannot provide comprehensive information, limiting the accuracy of behavioral analysis and health diagnosis.

[0006] This problem further posed challenges for dynamic behavior recognition. Ducks exhibit diverse and rapidly changing behavior patterns, and traditional models are limited in their ability to track and extract features in densely packed scenarios, making it difficult to achieve high-precision, real-time analysis.

[0007] A deeper challenge stems from the balance between data processing and system response speed. Edge devices need to run complex algorithms while maintaining low power consumption and low latency, which places higher demands on model optimization and system architecture design.

[0008] Therefore, how to achieve effective fusion of multimodal data, accurate behavior recognition, and low-power and high-efficiency edge computing in complex breeding environments has become a key issue in promoting the development of intelligent monitoring in duck houses. Summary of the Invention

[0009] The present invention provides an intelligent monitoring method and system for duck houses based on image recognition, aiming to solve at least one defect of the above-mentioned prior art.

[0010] One aspect of the present invention relates to an intelligent monitoring method for duck houses based on image recognition, comprising the following steps: Acquire environmental monitoring data from an environmental monitoring device and acquire original image data of the breeding area from an image acquisition device, wherein the environmental monitoring data includes temperature data, humidity data, and light intensity data; The original image data is processed using an image enhancement algorithm to obtain a preprocessed image data set. The image enhancement algorithm includes an illumination compensation module and an adaptive denoising module. According to the weight parameters output by the duck flock behavior and environment association model, the environmental monitoring data and the pre-processed image dataset are weightedly fused to generate a multimodal feature dataset; Determine whether the group behavior indicators in the multimodal feature dataset deviate from the preset threshold range. If the group behavior indicators in the multimodal feature dataset deviate from the preset threshold range, extract the spatiotemporal feature map through the convolutional neural network and combine it with the multi-target tracking algorithm to generate individual behavior trajectory data; Individual behavior trajectory data is input into a lightweight neural network model, which outputs health status classification results. The lightweight neural network model is deployed in edge computing devices. Generate environmental control instructions based on the health status classification results. The environmental control instructions include temperature compensation, humidity adjustment parameters, and ventilation intensity indicators; The environmental control instructions are transmitted to the environmental adjustment device to complete the dynamic environmental optimization of the breeding area.

[0011] Furthermore, the steps of obtaining environmental monitoring data from the environmental monitoring device and obtaining original image data of the breeding area from the image acquisition device include: Obtain temperature data, humidity data, and light intensity data from environmental monitoring devices, remove outliers using data cleaning tools, and obtain environmental monitoring data; The original image data of the breeding area is obtained from the image acquisition device.

[0012] Furthermore, the steps of using an image enhancement algorithm to process the original image data to obtain a preprocessed image data set include: The original image data of the breeding area is obtained from the image acquisition device, and the original image data is processed using the illumination compensation module to generate an illumination-balanced image by adjusting the pixel brightness value; For the illumination-balanced image, the adaptive denoising module is used to process the illumination-balanced image, and the noise is removed by the median filtering method to generate a denoised image; According to the denoised image, the edge detection algorithm is used to extract the target contour features in the denoised image. If the clarity of the target contour is greater than the preset threshold, the target contour features are retained and a feature image is generated; Through the feature image, the image segmentation algorithm is used to divide the breeding area in the feature image, generate a monitoring image containing the target area, and obtain a preprocessed image data set.

[0013] Furthermore, the steps of weighted fusion of the environmental monitoring data and the pre-processed image dataset based on the weight parameters output by the duck flock behavior and environment association model to generate a multimodal feature dataset include: The environmental data of temperature, humidity, and ammonia concentration in the breeding area are obtained from environmental sensors. The environmental data are processed using time series analysis methods, and the time series characteristics of each environmental factor are calculated using a sliding window method to obtain an environmental characteristic data set. According to the environmental feature dataset and the preprocessed image dataset, the weight parameters are used to perform weighted fusion on the environmental feature dataset and the preprocessed image dataset, and the initial multimodal feature set is generated by the matrix weighted summation method to obtain the fused feature set; For the fused feature set, the principal component analysis algorithm is used to reduce the dimension of the fused feature set. If the feature dimension after dimensionality reduction is less than the preset threshold, the reduced dimension feature is retained to obtain the optimized feature data set. By optimizing the feature data set, cluster analysis method is used to classify the optimized feature data set, and the monitoring area is divided according to the spatial distribution characteristics of duck flock behavior to obtain a multimodal feature data set.

[0014] Furthermore, determining whether the group behavior indicators in the multimodal feature dataset deviate from a preset threshold range; if the group behavior indicators in the multimodal feature dataset deviate from the preset threshold range, extracting a spatiotemporal feature graph through a convolutional neural network and combining it with a multi-target tracking algorithm to generate individual behavior trajectory data includes the following steps: Obtain group behavior indicators from a multimodal feature dataset, compare the group behavior indicators with a preset threshold range, determine whether the indicators deviate from the threshold range, and obtain a deviation state; If the deviation state is true, the time series data in the multimodal feature dataset is processed using a convolutional neural network to generate a spatiotemporal feature map; According to the spatiotemporal feature graph, a multi-target tracking algorithm is used to track individuals in group behavior and generate individual behavior trajectory data.

[0015] Furthermore, the individual behavior trajectory data is input into the lightweight neural network model to output the health status classification result. The steps of deploying the lightweight neural network model in the edge computing device include: Obtain the time series data from the individual behavior trajectory dataset, use the filtering algorithm to denoise the time series data, and obtain smooth trajectory data; If the time span of the smooth trajectory data exceeds the preset threshold, the frequency features of the smooth trajectory data are extracted by Fourier transform to obtain a frequency feature set; According to the frequency feature set, a lightweight neural network is used to classify the frequency feature set to obtain health status classification data; The health status classification data is transmitted to the dynamic monitoring module through the message queue in the edge computing device to determine whether the health status is abnormal.

[0016] Furthermore, the steps of generating an environmental control instruction according to the health status classification result, wherein the environmental control instruction includes a temperature compensation amount, a humidity adjustment parameter, and a ventilation intensity index include: Obtain an abnormal state identifier from the health state classification result, and use a preset mapping table to generate instructions including temperature compensation, humidity adjustment parameters, and ventilation intensity indicators to obtain an initial control instruction set; Based on the initial control instruction set and combined with the real-time sensor data in the edge computing device, a weighted average algorithm is used to adjust the temperature compensation, humidity adjustment parameters, and ventilation intensity indicators to obtain an optimized control instruction set. If the temperature compensation amount, humidity adjustment parameter or ventilation intensity index in the optimized control instruction set exceeds the preset threshold, the over-threshold parameter is calibrated by the fuzzy logic algorithm to obtain a calibrated control instruction set; The calibration control instruction set is transmitted to the environmental control module through the message queue in the edge computing device to determine the environmental control instructions.

[0017] Furthermore, the steps of transmitting the environmental control instructions to the environmental adjustment device to complete the dynamic environmental optimization of the breeding area include: Obtain environmental control instructions from the edge computing device, transmit them to the environmental control device through the message queue protocol, and confirm that the transmission of the environmental control instructions is completed; According to the temperature compensation, humidity adjustment parameters and ventilation intensity index in the environmental control instruction, the preset mapping table is used to adjust the operating state of the environmental control device to obtain the environmental control result; The real-time environmental data of the breeding area is obtained through the built-in sensors of the environmental control device. The weighted average algorithm is used to compare the real-time environmental data with the control parameters to determine whether the dynamic optimization meets the standards. If the dynamic optimization fails to meet the standards, the control parameters are recalibrated using a fuzzy logic algorithm to obtain an updated control instruction set and transmit it to the environmental control device to confirm that the optimization control is completed.

[0018] Another aspect of the present invention relates to an intelligent monitoring system for duck houses based on image recognition, which is used to implement the above-mentioned intelligent monitoring method for duck houses based on image recognition. The intelligent monitoring system for duck houses based on image recognition includes: A first acquisition module is used to acquire environmental monitoring data from an environmental monitoring device and to acquire original image data of a breeding area from an image acquisition device. The environmental monitoring data includes temperature data, humidity data, and light intensity data. The second acquisition module is used to process the original image data using an image enhancement algorithm to obtain a preprocessed image data set. The image enhancement algorithm includes an illumination compensation module and an adaptive denoising module; The first generation module is used to perform weighted fusion of the environmental monitoring data and the pre-processed image dataset based on the weight parameters output by the duck flock behavior and environment association model to generate a multimodal feature dataset; The second generation module is used to determine whether the group behavior indicators in the multimodal feature dataset deviate from the preset threshold range. If the group behavior indicators in the multimodal feature dataset deviate from the preset threshold range, the spatiotemporal feature graph is extracted through a convolutional neural network and combined with a multi-target tracking algorithm to generate individual behavior trajectory data; The output module is used to input individual behavior trajectory data into a lightweight neural network model and output health status classification results. The lightweight neural network model is deployed in the edge computing device; The third generation module is used to generate environmental control instructions based on the health status classification results. The environmental control instructions include temperature compensation, humidity adjustment parameters and ventilation intensity indicators; The optimization module transmits the environmental control instructions to the environmental adjustment device to complete the dynamic environmental optimization of the breeding area.

[0019] Furthermore, the first acquisition module includes: A first acquisition unit is configured to acquire temperature data, humidity data, and light intensity data from an environmental monitoring device, and remove abnormal values using a data cleaning tool to obtain environmental monitoring data; The second acquisition unit is used to acquire original image data of the breeding area from the image acquisition device.

[0020] The beneficial effects achieved by the present invention are: The present invention provides an intelligent monitoring method and system for duck houses based on image recognition. The environmental data and image data of the breeding area are obtained through an environmental monitoring device and an image acquisition device. The original image is pre-processed using an image enhancement algorithm, and multi-source data is fused according to a duck flock behavior and environment association model to generate a multimodal feature data set. When the group behavior indicators are abnormal, the present invention uses a convolutional neural network and a multi-target tracking algorithm to extract individual behavior trajectories, and classifies the health status through a lightweight neural network model deployed on the edge device. Finally, environmental control instructions are generated based on the classification results to achieve dynamic environmental optimization of the breeding area. Through multimodal data analysis and edge intelligence technology, the present invention realizes precise monitoring and intelligent adjustment of the duck breeding environment, effectively improving the breeding efficiency and the health level of the duck flock. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1The figure is a flow chart of an embodiment of an intelligent monitoring method for duck houses based on image recognition according to the present invention. DETAILED DESCRIPTION

[0022] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0023] like Figure 1 As shown, the first embodiment of the present invention proposes an intelligent monitoring method for duck houses based on image recognition, comprising the following steps: Step S100: Acquire environmental monitoring data from an environmental monitoring device, and acquire original image data of a breeding area from an image acquisition device. The environmental monitoring data includes temperature data, humidity data, and light intensity data.

[0024] Environmental monitoring devices refer to specialized equipment or systems used to detect, analyze, and record specific parameters or pollutant concentrations in the environment in real time. Their core function is to assess environmental quality and provide a scientific basis for management decisions.

[0025] An image acquisition device refers to a hardware device that converts actual scenes into digital images or videos through optical or electronic technology. Its core function is to capture static or dynamic visual information for subsequent processing and analysis.

[0026] Environmental monitoring data refers to a collection of quantifiable and analyzable digital information generated through the systematic observation, measurement, and recording of environmental elements and pollutants using scientific methods. Its core function is to objectively reflect the current status of environmental quality and its dynamic changes.

[0027] The original image data of the breeding area refers to the collection of visual information of the breeding scene directly obtained through image acquisition equipment (such as cameras and drones) without algorithm optimization or human intervention. Its core characteristics are originality and integrity.

[0028] Step S200: Processing the original image data using an image enhancement algorithm to obtain a pre-processed image data set. The image enhancement algorithm includes an illumination compensation module and an adaptive denoising module.

[0029] Image enhancement algorithms are a type of digital image processing technology that improves the visual quality of images or highlights key features through mathematical or computational models. Their core goal is to improve the usability of images to meet specific analysis or observation needs.

[0030] The illumination compensation module is a functional unit integrated into the image processing system. It uses algorithms to adjust the image's brightness distribution and color balance to eliminate visual distortion caused by uneven lighting, device deviations, or environmental interference. Its core goal is to enhance the visibility of key image areas and restore true colors, providing standardized input for subsequent analysis and recognition tasks.

[0031] The adaptive denoising module is an image processing component that dynamically adjusts denoising strategies and parameters. It automatically optimizes denoising by analyzing the noise characteristics of the input image (such as type, intensity, and distribution) to strike a balance between noise suppression and detail preservation. Its core goal is to eliminate noise interference in complex scenes, improve the image signal-to-noise ratio (SNR), and provide high-quality input for subsequent analysis tasks.

[0032] A preprocessed image dataset refers to a structured data set that is made suitable for machine learning model training by performing standardization and feature optimization on raw image data. Its core goal is to eliminate data noise, enhance key information, and unify data formats to improve the model's generalization ability and training efficiency.

[0033] Step S300: Perform weighted fusion on the environmental monitoring data and the pre-processed image dataset according to the weight parameters output by the duck flock behavior and environment association model to generate a multimodal feature dataset.

[0034] Duck flocking behavior refers to the group activity pattern formed by ducks due to survival needs and ecological adaptability, which is manifested in the coordinated actions among individuals to achieve functional goals such as improving foraging efficiency, defending against natural enemies, and selecting habitats.

[0035] The environmental association model is a theoretical framework that describes the dynamic interactive relationship between biological behavior and the living environment. It emphasizes that organisms achieve adaptive fit with the environment through behavioral adjustments to improve their survival and reproductive success rates.

[0036] Weight parameters are adjustable variables used in machine learning or statistical models to quantify the influence of input features on the output. Optimizing these parameters enables the model to capture underlying patterns in the data and achieve prediction or classification goals. Essentially, they are the knowledge carriers that the model learns from the data and determine the importance of different features in the decision-making process.

[0037] A multimodal feature dataset refers to a structured data set that contains two or more heterogeneous modal data (such as vision, text, audio, etc.), with semantic associations and spatiotemporal alignment relationships between the modalities. Its core goal is to enhance the artificial intelligence model's understanding and reasoning ability of the real world through the complementarity of multi-source heterogeneous data.

[0038] Step S400: Determine whether the group behavior indicators in the multimodal feature data set deviate from the preset threshold range. If the group behavior indicators in the multimodal feature data set deviate from the preset threshold range, extract the spatiotemporal feature map through the convolutional neural network and generate individual behavior trajectory data in combination with the multi-target tracking algorithm.

[0039] Swarm behavior indicators are a system of observational parameters that quantify the interaction patterns and collaborative efficiency of biological groups in specific environments. They are used to characterize group dynamics, assess group health, and predict ecological adaptability. Their core function is to reveal the internal organizational rules and external environmental response mechanisms of a group through measurable behavioral data.

[0040] The Convolutional Neural Network (CNN) is a deep learning architecture designed specifically for processing grid-topological data (such as images, videos, and speech). It automatically extracts spatial or temporal features through local perception, weight sharing, and hierarchical abstraction. Its core concept is to mimic the hierarchical information processing of biological visual systems, enabling progressive feature learning from low-level textures to high-level semantics.

[0041] A spatiotemporal feature graph is a high-dimensional data structure that integrates time and space. By dynamically capturing the evolution of spatiotemporal correlations between nodes, it enables joint modeling of spatiotemporal dependencies in complex systems. Essentially, it integrates temporal changes and spatial topology into a unified representation, supporting dynamic system analysis tasks such as behavior prediction and anomaly detection.

[0042] Multi-Object Tracking (MOT) is a computer vision technology that continuously tracks the positions, motion states, and identities of multiple dynamic objects in a video sequence through object detection, data association, and trajectory management. Its core goal is to ensure trajectory continuity and maintain identity consistency even in scenarios where the number of objects changes dynamically (such as when new objects appear, disappear, or become obscured).

[0043] Individual behavior trajectory data is a highly granular data set that continuously samples and records an individual's dynamic activities in physical or digital space, encompassing multiple dimensions such as time, space, and behavioral attributes. Essentially, it represents a continuous representation of individual behavior patterns using time and space as coordinate axes, supporting application scenarios such as behavior prediction, privacy protection, and resource optimization. In this embodiment, individual behavior trajectory data is used to output health status classification results.

[0044] Step S500: Input individual behavior trajectory data into a lightweight neural network model, output health status classification results, and deploy the lightweight neural network model in an edge computing device.

[0045] A lightweight neural network model is a low-power deep learning architecture that uses structural optimization and compression techniques to reduce parameter count and computational complexity, making it suitable for resource-constrained scenarios such as mobile terminals and embedded devices. Its core goal is to achieve efficient model deployment and real-time inference while ensuring task performance, addressing the pain point of traditional deep neural networks' excessive dependence on hardware resources.

[0046] The Health Status Classification is a comprehensive assessment system for an individual's physical, psychological, and social adaptability. It aims to quantify health risks and guide health management strategies through grading criteria. Its classification criteria cover functional status, disease status, and social adaptability, forming a multi-dimensional, dynamic assessment framework.

[0047] Edge computing devices are distributed intelligent hardware units deployed near data sources or end users. They integrate computing, storage, and networking capabilities to perform local data processing and analysis. Their core function is to replace cloud-based backhaul with local computing, reducing response latency and improving data privacy. This makes them suitable for real-time scenarios such as the Internet of Things and industrial automation.

[0048] Step S600: Generate an environmental control instruction according to the health status classification result. The environmental control instruction includes a temperature compensation amount, a humidity adjustment parameter, and a ventilation intensity index.

[0049] Environmental control instructions are a set of digital commands that dynamically adjust specific environmental factors (such as temperature, humidity, and air quality) through preset rules or manually set parameters. Essentially, they achieve precise control and optimization of physical or virtual environmental conditions through the integration of sensor monitoring and actuators.

[0050] Step S700: Transmit the environmental control instruction to the environmental adjustment device to complete the dynamic environmental optimization of the breeding area.

[0051] An environmental conditioning device is a physical device or system that dynamically adjusts specific environmental parameters (such as temperature, humidity, light intensity, and gas concentration) through sensor perception, controller decision-making, and actuator linkage. Its core function is to maintain environmental parameters within preset ranges through closed-loop control strategies, adapting to the needs of various fields such as industrial production, agricultural planting, and building energy conservation. In this embodiment, the environmental conditioning device is used to regulate the dynamic environment of a farming area.

[0052] Dynamic environmental optimization of aquaculture areas involves a systematic project that continuously adjusts environmental parameters within the aquaculture area (such as water quality, temperature, humidity, and gas concentrations) through real-time monitoring and intelligent control technologies to improve resource utilization efficiency, ensure animal health, and achieve sustainable ecological development. Its core goal is to establish a closed-loop management mechanism of "monitoring-analysis-control," balancing production efficiency and environmental carrying capacity.

[0053] Furthermore, the intelligent duck house monitoring method based on image recognition provided in this embodiment includes step S100: Step S110: Acquire temperature data, humidity data, and light intensity data from an environmental monitoring device, remove abnormal values using a data cleaning tool, and obtain environmental monitoring data.

[0054] Environmental monitoring data is set to a temperature range of 10-35°C, a humidity range of 20-80%, and a light intensity of 0-100,000 lux. A temperature reading of 50°C is clearly an outlier and needs to be removed using a data cleaning tool.

[0055] Cleaning methods can use statistical outlier detection, such as using the mean and standard deviation to remove data exceeding twice the standard deviation. This approach effectively filters out erroneous data caused by sensor failure or external interference, ensuring that subsequent analysis is based on accurate data and improving the reliability of aquaculture environment control.

[0056] Specifically, data cleaning tools can be further optimized based on time series analysis. For example, within a farm, temperature data fluctuates between 22-25°C over the course of an hour, and then a reading of 38°C suddenly appears. Using time series smoothing algorithms, such as moving averages, this value is identified as deviating from the trend, marked as an anomaly, and replaced with the average of the surrounding data. This not only preserves data continuity but also prevents outliers from interfering with environmental trend analysis, facilitating precise temperature and humidity control and ensuring a stable growth environment for livestock or poultry.

[0057] Step S120: Acquire original image data of the breeding area from the image acquisition device.

[0058] In one possible implementation, image acquisition devices are deployed at key locations in the farming area, such as feed troughs or rest areas, to capture high-resolution image data. For example, a camera could capture images once an hour to record livestock activity or environmental conditions.

[0059] Raw images may have quality issues due to insufficient lighting or lens stains, requiring preprocessing to ensure data usability. Image preprocessing can include brightness adjustment and noise filtering, such as enhancing the contrast of low-light images through histogram equalization or applying Gaussian filtering to remove lens smudge noise. These processes ensure image clarity, facilitating subsequent analysis of livestock behavior or environmental hygiene, and improving livestock management efficiency. For example, combining environmental data with image data allows for comprehensive analysis. A farm discovered that nighttime light intensity was below 500 lux, and images showed reduced livestock activity. Combined with temperature data of 20°C and humidity of 60%, it was speculated that insufficient light may have affected livestock feeding.

[0060] Based on this, the lighting in the farm was adjusted to 1000 lux, and subsequent image data showed that livestock activity had returned to normal. This data-driven analysis can accurately pinpoint environmental issues, optimize farming conditions, and improve production and animal welfare.

[0061] It's important to note that automated data cleaning and image preprocessing rely on the IoT platform. For example, sensor and camera data are uploaded to the cloud, where cleaning tools address outliers in real time and image processing algorithms automatically optimize image quality. The processed data can then be used to generate visual reports for farmers' reference. This automated process reduces manual intervention, improves data processing efficiency, ensures real-time and accurate environmental monitoring, and provides reliable support for scientific farming.

[0062] In one possible implementation, cleaned environmental data and image data can be used in predictive models. For example, based on historical temperature, humidity, and light data and images of livestock activity, the impact of environmental changes on livestock health can be predicted. This predictive model can provide early warning of potential risks, such as diseases caused by high temperatures and humidity, prompting farmers to implement cooling or ventilation measures, thereby reducing economic losses and improving the level of intelligent farming.

[0063] Furthermore, the intelligent duck house monitoring method based on image recognition provided in this embodiment includes step S200: Step S210: Acquire original image data of the breeding area from the image acquisition device, process the original image data using the illumination compensation module, and generate an illumination balanced image by adjusting the pixel brightness value.

[0064] In the aquaculture environment monitoring scenario, the original image data of the aquaculture area is obtained from the image acquisition device, and a preprocessed image data set is generated through a series of image processing technologies, involving technical topics such as illumination compensation, adaptive denoising, edge detection and image segmentation.

[0065] The following analysis and examples focus on these topics, focusing on aquaculture environment monitoring and highlighting the technical implementation methods and logical progression. For example, the light compensation module generates a light-balanced image by adjusting pixel brightness values.

[0066] Lighting conditions in farms often fluctuate over time or due to weather, resulting in uneven image brightness and affecting subsequent analysis. Light compensation analyzes the image's brightness distribution, enhancing dark details and suppressing overly bright areas.

[0067] In one implementation, a camera is deployed above a feed trough on a farm. Due to insufficient light on a cloudy day, the image is generally dark. The illumination compensation module adjusts pixel values based on the brightness histogram, raising low-brightness pixels to an appropriate range. For example, by increasing the grayscale value from 50 to 100, the module generates a balanced image. This processing ensures clear image details, facilitating subsequent feature extraction.

[0068] Step S220: Process the illumination-balanced image using an adaptive denoising module, remove noise using a median filtering method, and generate a denoised image.

[0069] For example, the adaptive denoising module uses median filtering to remove noise and generate a denoised image. Farm images often introduce noise due to dust or lens stains, which degrades image quality. Median filtering effectively removes isolated noise points by replacing pixel values with the median of their neighboring pixels.

[0070] In one implementation, a camera image of a livestock resting area contains white noise due to dust. The adaptive denoising module analyzes the image noise distribution and applies a median filter with a 3x3 pixel window to replace the grayscale value of the noise point with the median value of the surrounding pixels. For example, an outlier value of 255 is replaced with 120. The denoised image retains the details of the livestock's outline, providing high-quality input for edge detection.

[0071] Step S230: Based on the denoised image, an edge detection algorithm is used to extract the target contour features in the denoised image. If the clarity of the target contour is greater than a preset threshold, the target contour features are retained to generate a feature image.

[0072] For example, edge detection algorithms extract the outline features of objects in a denoised image, retaining only those whose clarity exceeds a preset threshold. Edge detection identifies changes in pixel grayscale to outline the boundaries of an object, such as livestock or a feed trough.

[0073] In one implementation, the Canny edge detection algorithm was used to process the denoised image, with a sharpness threshold of 0.8. The livestock outline in the image had a high grayscale gradient and a sharpness of 0.9, so it was retained as a feature. However, the background debris outline had a sharpness of only 0.6 and was removed. The resulting feature image emphasized the livestock shape, facilitating region segmentation.

[0074] Step S240: Using the feature image, an image segmentation algorithm is used to divide the breeding area in the feature image, generate a monitoring image containing the target area, and obtain a pre-processed image data set.

[0075] For example, an image segmentation algorithm divides the livestock area in a feature image into sections and generates a surveillance image containing the target area. Image segmentation separates the target area from the background through pixel classification, such as distinguishing livestock from the ground.

[0076] In one implementation, a region-growing-based segmentation algorithm is used, selecting pixels within the livestock's outline as seed points and expanding the segmentation to encompass regions of similar grayscale. After segmentation, the surveillance image only contains the area where the livestock is located, excluding irrelevant background. The final preprocessed image dataset integrates multiple frames of surveillance imagery, providing reliable data for subsequent livestock behavior analysis.

[0077] It should be noted that the above technical process can be automated through the Internet of Things platform. The image acquisition device uploads raw data to the cloud, where it is processed sequentially by the illumination compensation, denoising, edge detection, and segmentation modules to generate a pre-processed image dataset. This automated process reduces manual intervention, improves data processing efficiency, and provides farms with real-time, accurate monitoring data, supporting scientific management decisions.

[0078] In one embodiment, integrating historical environmental data can expand solution functionality. For example, analyzing light-balanced images and temperature data can determine whether livestock activity is affected by the environment. For example, if nighttime images show livestock congregating in a corner, the environmental monitoring device can combine this data with temperature data to obtain temperature, humidity, and light intensity data. This data can then be combined with the image acquisition device to obtain raw image data of the breeding area. Data cleaning is then required to remove outliers and ensure data reliability.

[0079] The following analysis and examples revolve around these technical themes, focusing on aquaculture environment monitoring scenarios and highlighting the implementation of the technologies and their beneficial effects.

[0080] For example, environmental monitoring devices are often deployed within farms, where sensors collect data in real time. Temperature sensors monitor air temperature, humidity sensors measure relative humidity, and light intensity sensors record light levels.

[0081] In one possible implementation, the device collects data at 28°C every minute, inferring that high temperatures are causing behavioral abnormalities. Based on the preprocessed image dataset, it further analyzes livestock distribution patterns and optimizes the ventilation layout of the farm. This multifaceted analysis supports each other, improving the precision of environmental control and providing technical support for intelligent farming.

[0082] Furthermore, the intelligent duck house monitoring method based on image recognition provided in this embodiment includes step S300: Step S310: Acquire environmental data of temperature, humidity, and ammonia concentration in the breeding area from environmental sensors, process the environmental data using a time series analysis method, and use a sliding window method to calculate the time series characteristics of each environmental factor to obtain an environmental characteristic data set.

[0083] In the aquaculture environment monitoring scenario, the fusion analysis of environmental sensors and image data is an important technical means to achieve intelligent management.

[0084] The following analysis and examples focus on technical topics such as time series analysis of temperature, humidity, and ammonia concentration, sliding window feature extraction, weighted fusion, principal component analysis dimensionality reduction, and cluster analysis. Focusing on monitoring duck behavior in a farm, the paper highlights the technical implementation methods and logical progression. For example, environmental sensors are deployed in a farm to collect temperature, humidity, and ammonia concentration data, forming a time series dataset.

[0085] Time series analysis uses a sliding window approach to extract dynamic features. In principle, a sliding window traverses the data at a fixed time step and calculates statistical indicators within the window, such as the mean and variance.

[0086] Preferably, the sensor records data every minute, with a sliding window of 30 minutes. The mean and variance of the temperature within the window are calculated. For example, if the temperature data within a window fluctuates between 25 and 27°C, with a mean of 26°C and a variance of 0.5, this indicates temperature stability. These features constitute an environmental feature dataset that captures patterns of environmental change.

[0087] Step S320: Based on the environmental feature dataset and the preprocessed image dataset, weight parameters are used to perform weighted fusion on the environmental feature dataset and the preprocessed image dataset, and an initial multimodal feature set is generated by a matrix weighted summation method to obtain a fused feature set.

[0088] For example, based on the environmental feature dataset and the preprocessed image dataset, weighted fusion is performed to generate an initial multimodal feature set. Weighted fusion balances the importance of different data sources through matrix weighted summation.

[0089] Specifically, environmental features include a mean temperature of 26°C, a mean humidity of 70%, and an ammonia concentration of 0.2 ppm; the image dataset also includes the distribution density of ducks. Weighting parameters are assigned based on the data's impact on duck behavior, such as a temperature weight of 0.4 and an image density weight of 0.5. After fusion, a comprehensive feature vector is generated, reflecting the correlation between the environment and behavior.

[0090] Step S330: For the fused feature set, a principal component analysis algorithm is used to perform dimensionality reduction processing on the fused feature set. If the feature dimension after dimensionality reduction is less than a preset threshold, the dimensionality reduction feature is retained to obtain an optimized feature data set.

[0091] For example, the principal component analysis algorithm reduces the dimensionality of the fused feature set, reducing computational complexity. Dimensionality reduction preserves key information and eliminates redundant dimensions.

[0092] Preferably, the fused feature set contains 10 features, with a preset threshold of 5 dimensions. Principal component analysis calculates the feature covariance matrix and extracts the top 5 dimensions with the highest contribution. For example, after dimensionality reduction, the combined features of temperature, humidity, and duck density are retained, reducing the dimension from 10 to 5, generating an optimized feature dataset for subsequent analysis.

[0093] Step S340 , by optimizing the feature data set, classifying the optimized feature data set using a cluster analysis method, dividing the monitoring area according to the spatial distribution characteristics of the duck flock behavior, and obtaining a multimodal feature data set.

[0094] For example, cluster analysis classifies the optimized feature dataset and divides the duck behavior areas. Clustering is based on spatial distribution characteristics, classifying similar behavior patterns. Specifically, K-means clustering is used, and three clusters are set, corresponding to the aggregation, dispersion, and stationary states of the ducks. In one embodiment, the optimized feature dataset shows that the temperature in a certain area is 28°C and the duck density is high. The clustering results indicate that the ducks gather near the ventilation holes due to the high temperature. The generated monitoring area image is annotated with the aggregation areas to support precise management.

[0095] It should be noted that the above process can be automated through the Internet of Things platform. Sensor and camera data are uploaded to the cloud, and time series analysis, fusion, dimensionality reduction, and clustering are performed in sequence to generate a multimodal feature dataset.

[0096] Ideally, historical data can be combined to expand functionality. For example, this can be used to analyze duck flocking patterns during high-temperature periods, optimize ventilation equipment layout, and improve the accuracy of environmental control. This multi-faceted analysis supports each other and provides a reliable data foundation for intelligent farming.

[0097] Furthermore, the intelligent duck house monitoring method based on image recognition provided in this embodiment includes step S400: Step S410: Obtain a group behavior index from the multimodal feature data set, compare the group behavior index with a preset threshold range, determine whether the index deviates from the threshold range, and obtain a deviation state.

[0098] In the duck flock behavior monitoring scenario in a farm, extracting group behavior indicators from multimodal feature data sets, judging deviation states, generating spatiotemporal feature maps, and tracking individual behavior trajectories are the core technical paths to achieve intelligent management.

[0099] The following analysis and examples center around these technical topics, focusing on monitoring duck flock behavior on farms, highlighting implementation methods and logical progression. For example, flock behavior metrics are extracted from multimodal feature datasets to quantify the collective behavior patterns of ducks. These metrics might include the average movement speed, aggregation density, and activity frequency of the flock.

[0100] In principle, by analyzing time series data and image data, statistical characteristics within a specific time period are calculated. Preferably, the sensor records the duck flock's position every minute. Combined with the distribution image captured by the camera, the average movement speed of the duck flock over a 30-minute period is calculated to be 0.5 meters per minute, and the flock density is 10 ducks per square meter.

[0101] Indicator extraction is automatically completed through cloud-based algorithms to ensure real-time performance. For example, it determines whether group behavior indicators deviate from the preset threshold range to identify abnormal behavior. The preset threshold is set based on historical data and farming experience. For example, the normal aggregation density is 8-12 ducks per square meter and the movement speed is 0.3-0.7 meters per minute. If the aggregation density reaches 15 ducks per square meter during a certain period, exceeding the upper threshold, the deviation status is true. It should be noted that the deviation status triggers subsequent analysis to locate the cause of the abnormality, such as high temperature causing the ducks to gather. The cloud system records the deviation status and generates an alarm signal.

[0102] Step S420: If the deviation state is true, a convolutional neural network is used to process the time series data in the multimodal feature dataset to generate a spatiotemporal feature map.

[0103] For example, for time series data where the deviation state is true, a convolutional neural network is used to generate a spatiotemporal feature map. The convolutional neural network extracts local patterns in the time series through convolution operations, such as the correlation between temperature changes and the movement of duck flocks.

[0104] Preferably, the input data is a 30-minute time series of temperature and location, and the network outputs a feature map that includes both the temporal dimension and the spatial distribution. In one embodiment, the feature map shows that during periods of high temperature, ducks move toward the farm, reflecting the spatial pattern of environmental influences.

[0105] Step S430: Based on the spatiotemporal feature graph, a multi-target tracking algorithm is used to track individuals in the group behavior to generate individual behavior trajectory data.

[0106] For example, based on the spatiotemporal feature map, a multi-target tracking algorithm is used to generate individual behavioral trajectory data. The multi-target tracking algorithm tracks the movement path of each duck by correlating the individual positions in the feature map.

[0107] Preferably, the algorithm updates the position once per second, generating a continuous trajectory. In one embodiment, tracking revealed that a particular duck frequently moved to a ventilation vent during periods of high temperatures. The trajectory data showed that the duck moved a distance of 10 meters, indicating the individual's sensitivity to environmental changes.

[0108] Trajectory data supports refined management, such as identifying abnormal individuals. It should be noted that the above process is automated through the IoT platform.

[0109] Sensor and camera data are uploaded in real time, and group behavior indicator extraction, deviation judgment, spatiotemporal feature map generation and individual tracking are performed sequentially.

[0110] Optimally, historical trajectory data can be combined to analyze the duck flock's adaptation to environmental changes, optimize farm equipment layout, and improve management efficiency. This multi-faceted analysis supports each other and provides a precise data foundation for intelligent farming.

[0111] Furthermore, the intelligent duck house monitoring method based on image recognition provided in this embodiment includes step S500: Step S510: Obtain time series data from the individual behavior trajectory dataset, and use a filtering algorithm to perform denoising on the time series data to obtain smooth trajectory data.

[0112] For example, in the context of monitoring duck flock behavior on a farm, obtaining time series data from individual behavioral trajectory datasets is essential for precise health management. Time series data typically records the location coordinates of each duck over a specific time period, such as those generated from motion paths captured by worn RFID tags or cameras.

[0113] In principle, the sensor records its position once per second, forming a data stream containing timestamps and spatial coordinates. In one example, the trajectory data for a particular duck over a 10-minute period includes 600 location points, reflecting its movement patterns within the farm.

[0114] It's important to note that raw data often contains noise due to sensor jitter or environmental interference, affecting the accuracy of subsequent analysis. For example, filtering algorithms can be used to denoise time series data to produce smooth trajectory data. Filtering algorithms, such as the Kalman filter, eliminate random fluctuations in position data through prediction and update steps.

[0115] Preferably, the algorithm processes 10 minutes of trajectory data with a one-second time step to generate a smooth motion curve. In one embodiment, the original trajectory of a duck shows brief position jumps, but after filtering, a continuous path is obtained, reflecting its true movement trend. It is important to note that smoothing the data preserves key characteristics of the behavioral pattern, facilitating subsequent frequency analysis.

[0116] Step S520: If the time span of the smoothed trajectory data exceeds a preset threshold, frequency features of the smoothed trajectory data are extracted by Fourier transform to obtain a frequency feature set.

[0117] For example, if the time span of the smoothed trajectory data exceeds a preset threshold, the frequency features are extracted through Fourier transform.

[0118] The time span threshold is set based on farming experience. For example, under normal circumstances, a duck's trajectory spans 5-15 minutes. If a duck's trajectory spans 20 minutes, frequency analysis is triggered. Fourier transform converts the time series into the frequency domain, extracting the frequency components of periodic behavior. In one embodiment, the transformation results show that the movement frequency of a duck is concentrated around 0.1 Hz, reflecting its regular gait. This frequency feature set provides a quantitative basis for subsequent classification.

[0119] Step S530: Based on the frequency feature set, a lightweight neural network is used to classify the frequency feature set to obtain health status classification data.

[0120] For example, a lightweight neural network can be used to classify frequency feature sets to obtain health status classification data. Lightweight neural networks such as Mobile Net use a shallow convolutional structure, making them suitable for edge device computing.

[0121] Preferably, the input is a vector containing frequency features, and the output is a classification label of healthy, subhealthy, or abnormal. In one embodiment, the frequency features of a particular duck showed an abnormal proportion of high-frequency movements, and the network classified it as subhealthy. It should be noted that the lightweight design ensures real-time performance and is suitable for edge computing scenarios in farms.

[0122] Step S540: Transmit the health status classification data to the dynamic monitoring module through the message queue in the edge computing device to determine whether the health status is abnormal.

[0123] For example, health status classification data can be transmitted to the dynamic monitoring module through the message queue in the edge computing device. Message queues such as the MQTT protocol support low-bandwidth and highly reliable transmission.

[0124] Preferably, the classified data is packaged and sent every minute, and the dynamic monitoring module receives it and determines whether its health status is abnormal. In one embodiment, after a duck is classified as abnormal, the module analyzes its trajectory based on historical data and finds that it has been staying in a corner for a long time, indicating a possible health problem.

[0125] It should be noted that the combination of edge computing and message queues enables efficient data flow and supports real-time monitoring. For example, the above process is automated through the IoT platform.

[0126] RFID tag and camera data are uploaded in real time, and filtering, Fourier transform, neural network classification and data transmission are performed sequentially.

[0127] Optimally, edge devices generate hourly health status reports, and the dynamic monitoring module combines environmental data to analyze the causes of abnormalities, such as high temperature or insufficient feed. These multiple technical links support each other, forming a complete chain from data collection to health assessment, providing precise support for intelligent farm management.

[0128] Furthermore, the intelligent duck house monitoring method based on image recognition provided in this embodiment includes step S600: Step S610: Obtain an abnormal state identifier from the health state classification result, and use a preset mapping table to generate instructions including a temperature compensation amount, a humidity adjustment parameter, and a ventilation intensity index to obtain an initial control instruction set.

[0129] In the scenario of duck health management in a farm, obtaining abnormal status identification from the health status classification results and generating environmental control instructions are key links in realizing intelligent farming.

[0130] The following analysis and examples illustrate each technical topic, focusing on a single scenario and closely linking it to the field of duck flock behavior monitoring based on historical data. For example, abnormal status identification within health status classification results is typically based on the output of a lightweight neural network, identifying abnormal behavior in a particular duck, such as prolonged periods of time spent in a corner. This abnormal status identification serves as a trigger signal, indicating that adjustments to the farm environment are necessary to improve duck flock health.

[0131] A pre-set mapping table correlates abnormal conditions with environmental parameters to generate an initial set of control instructions. Designed based on farming experience, the mapping table includes temperature compensation, humidity adjustment parameters, and ventilation intensity indicators. For example, if the abnormal condition is identified as "low activity," the mapping table might specify a temperature compensation increase of 2 degrees Celsius, a humidity adjustment increase of 5%, and a medium ventilation intensity indicator. This mapping ensures that instructions directly correspond to abnormal conditions, providing a foundation for subsequent optimization.

[0132] Step S620: Based on the initial control instruction set and in combination with the real-time sensor data in the edge computing device, a weighted average algorithm is used to adjust the temperature compensation amount, humidity adjustment parameters, and ventilation intensity index to obtain an optimized control instruction set.

[0133] In one embodiment, after the initial control instruction set is generated, it is optimized based on real-time sensor data from edge computing devices. This sensor data includes temperature, humidity, and air velocity within the farm. For example, the current temperature is 22 degrees Celsius, the humidity is 60%, and the ventilation intensity is low.

[0134] The weighted average algorithm adjusts parameters based on sensor data and the initial command set, with weights determined by the environmental impact on the ducks' health. Temperature might have a weight of 0.5, humidity 0.3, and ventilation 0.2. If the initial command calls for a temperature increase of 24°C, the algorithm, taking into account the current temperature of 22°C, calculates an adjusted temperature increase of 1.8°C. This optimized control command set reflects the balance between the real-time environment and the target parameters, improving control accuracy.

[0135] Step S630: If the temperature compensation amount, humidity adjustment parameter or ventilation intensity index in the optimization control instruction set exceeds the preset threshold, the exceeding threshold parameters are calibrated by a fuzzy logic algorithm to obtain a calibration control instruction set.

[0136] For example, if a parameter in the optimized control instruction set exceeds a preset threshold, such as a temperature offset exceeding 3 degrees Celsius, a fuzzy logic algorithm initiates a calibration. Fuzzy logic handles complex environmental variables by defining fuzzy sets (e.g., "temperature is too high," "humidity is appropriate") and a rule base (e.g., "if temperature is too high and humidity is low, reduce temperature offset"). After calibration, the temperature offset might be adjusted to 2.5 degrees Celsius, ensuring the instruction is within a safe range. The flexibility of fuzzy logic is well-suited to multi-variable farm scenarios, ensuring the adaptability of control instructions.

[0137] Step S640: Transmit the calibration control instruction set to the environmental control module through the message queue in the edge computing device to determine the environmental control instruction.

[0138] In one embodiment, the calibration control instruction set is transmitted to the environmental control module via a message queue.

[0139] The message queue uses the MQTT protocol, ensuring efficient transmission even under low bandwidth conditions. For example, a command set containing temperature, humidity, and ventilation parameters is sent in a package every 30 seconds. Upon receiving this command, the environmental control module activates the heater, humidifier, or fan to implement the control. For example, if a farm receives a command to increase the temperature by 2.5 degrees Celsius, the heater starts, and after 10 minutes, the temperature stabilizes at the target value. The real-time and reliability of the message queue ensures rapid command execution and timely environmental control.

[0140] It's important to note that the above process is automated through edge computing devices. Sensor data is collected in real time, and command generation, optimization, calibration, and transmission are seamlessly integrated. For example, a farm detected several ducks staying abnormally still. By optimizing the command set to adjust temperature and ventilation, the ducks' activity levels returned to normal. This complete chain, from abnormality identification to environmental control, provides precise support for farms and improves the efficiency of duck health management.

[0141] Furthermore, the intelligent duck house monitoring method based on image recognition provided in this embodiment includes step S700: Step S710: Obtain an environmental control instruction from the edge computing device, transmit it to the environmental adjustment device through the message queue protocol, and determine that the transmission of the environmental control instruction is completed.

[0142] For example, in the health management scenario of a duck farm, edge computing devices transmit environmental control instructions to environmental conditioning devices through a message queue protocol to ensure efficient delivery of the instructions.

[0143] As the data processing core, the edge computing device generates an instruction set containing temperature compensation, humidity adjustment parameters and ventilation intensity indicators in real time.

[0144] Using the MQTT (Message Queuing Telemetry Transport) protocol, commands are packaged and transmitted every 30 seconds, including commands such as increasing the temperature by 2 degrees Celsius, increasing the humidity by 4%, and adjusting the ventilation intensity to medium. Upon receiving the command, the environmental control device immediately parses and confirms the completion of the transmission. The lightweight nature of the MQTT protocol ensures stable command transmission even under low bandwidth conditions, making it particularly suitable for farm network environments.

[0145] Step S720: According to the temperature compensation amount, humidity adjustment parameter and ventilation intensity index in the environmental control instruction, the preset mapping table is used to adjust the operating state of the environmental control device to obtain the environmental control result.

[0146] Specifically, the environmental conditioning device adjusts its operating state according to a preset mapping table. This mapping table is designed based on the ducks' sensitivity to the environment. For example, a temperature compensation of 2 degrees Celsius corresponds to the heater turning on to medium, a 4% increase in humidity triggers the humidifier to run at low speed, and medium ventilation intensity activates the fan at 50% power. After the device is operational, the environmental conditioning results in a stable temperature of 24 degrees Celsius in the breeding area, a humidity of 64%, and a moderate air flow rate. The pre-set logic of the mapping table ensures that commands are precisely matched to the device's operating state, quickly responding to the ducks' health needs.

[0147] Step S730: Acquire real-time environmental data of the breeding area through the built-in sensor of the environmental adjustment device, use a weighted average algorithm to compare the real-time environmental data with the control parameters, and determine whether the dynamic optimization meets the standards.

[0148] In one embodiment, the environmental conditioning device has built-in sensors that collect real-time environmental data of the farming area, including temperature, humidity, and air velocity. For example, the sensors detect that the current temperature is 23 degrees Celsius, the humidity is 62%, and the ventilation intensity is low.

[0149] A weighted average algorithm compares real-time data with control parameters, assigning weights of 0.5 for temperature, 0.3 for humidity, and 0.2 for ventilation. If the target temperature is 24°C, the algorithm calculates the deviation and determines whether dynamic optimization meets the target. If the deviation is less than a preset threshold, such as 0.5°C, optimization meets the target. If the deviation is larger, such as a temperature of only 22.5°C, the target is considered not met, triggering a second calibration.

[0150] Step S740: If the dynamic optimization fails to meet the requirements, the control parameters are recalibrated using a fuzzy logic algorithm to obtain an updated control instruction set and transmit it to the environmental control device to determine that the optimization control is completed.

[0151] For example, if dynamic optimization fails to meet targets, the fuzzy logic algorithm performs a second calibration of the control parameters. Fuzzy logic defines fuzzy sets, such as "temperature is low" and "humidity is appropriate," and combines them with a rule base, such as "if temperature is low and ventilation is insufficient, increase temperature compensation and increase ventilation," to generate an updated control instruction set. For example, after calibration, the temperature compensation is adjusted to 2.2 degrees Celsius, and the ventilation intensity is increased to 60% power.

[0152] The updated instruction set is then transmitted to the environmental control device via a message queue. The device then adjusts the heater and fan operating status accordingly. After 10 minutes, the temperature stabilizes at 24 degrees Celsius, completing the optimized control. The flexibility of fuzzy logic adapts to the multivariable environment of the farm, ensuring precise control.

[0153] It is understandable that the above process ensures the real-time and adaptability of environmental adjustment through the collaboration of technologies such as edge computing, message queues, and fuzzy logic.

[0154] A dynamic optimization mechanism driven by sensor data, combined with preset mapping tables and calibration algorithms, enables environmental control devices to rapidly respond to changes in duck flock health needs. For example, a drop in nighttime temperature at a farm triggered an adjustment command. After sensor data guided the calibration, the environment quickly returned to a suitable state, and ducks resumed normal activity. The entire solution, from command transmission to optimized control, forms a closed loop, providing efficient support for intelligent farming.

[0155] The present invention relates to an intelligent monitoring system for duck houses based on image recognition, which is used to implement the above-mentioned intelligent monitoring method for duck houses based on image recognition. The intelligent monitoring system for duck houses based on image recognition includes a first acquisition module, a second acquisition module, a first generation module, a second generation module, an output module, a third generation module and an optimization module, wherein the first acquisition module is used to obtain environmental monitoring data from an environmental monitoring device and obtain original image data of a breeding area from an image acquisition device, and the environmental monitoring data includes temperature data, humidity data and light intensity data; the second acquisition module is used to process the original image data using an image enhancement algorithm to obtain a preprocessed image data set, and the image enhancement algorithm includes an illumination compensation module and an adaptive denoising module; the first generation module is used to process the environmental monitoring data according to the weight parameters output by a model of association between duck flock behavior and environment The data is weightedly fused with the preprocessed image data set to generate a multimodal feature data set; the second generation module is used to determine whether the group behavior indicators in the multimodal feature data set deviate from the preset threshold range. If the group behavior indicators in the multimodal feature data set deviate from the preset threshold range, the spatiotemporal feature map is extracted through the convolutional neural network, and the individual behavior trajectory data is generated in combination with the multi-target tracking algorithm; the output module is used to input the individual behavior trajectory data into the lightweight neural network model and output the health status classification result. The lightweight neural network model is deployed in the edge computing device; the third generation module is used to generate environmental control instructions according to the health status classification result. The environmental control instructions include temperature compensation, humidity adjustment parameters and ventilation intensity indicators; the optimization module transmits the environmental control instructions to the environmental adjustment device to complete the dynamic environment optimization of the breeding area.

[0156] Furthermore, the intelligent duck house monitoring system based on image recognition provided by this embodiment, the first acquisition module includes a first acquisition unit and a second acquisition unit, wherein the first acquisition unit is used to obtain temperature data, humidity data and light intensity data from the environmental monitoring device, and remove outliers through a data cleaning tool to obtain environmental monitoring data; the second acquisition unit is used to obtain original image data of the breeding area from the image acquisition device.

[0157] Compared with the existing technology, the intelligent duck house monitoring method and system based on image recognition provided by this embodiment obtains environmental data and image data of the breeding area through environmental monitoring devices and image acquisition devices, pre-processes the original image using image enhancement algorithms, and fuses multi-source data according to the duck group behavior and environment association model to generate a multimodal feature data set. When the group behavior indicators are abnormal, this embodiment uses convolutional neural networks and multi-target tracking algorithms to extract individual behavior trajectories, and classifies the health status through a lightweight neural network model deployed on the edge device. Finally, environmental control instructions are generated based on the classification results to achieve dynamic environmental optimization of the breeding area. Through multimodal data analysis and edge intelligence technology, this embodiment achieves precise monitoring and intelligent adjustment of the duck breeding environment, effectively improving the breeding efficiency and the health level of the duck group.

[0158] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they are aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the invention. Thus, the present invention is intended to include such changes and modifications as fall within the scope of the claims and their equivalents.

Claims

1. An intelligent monitoring method for duck houses based on image recognition, characterized in that: The following steps are involved: Acquire environmental monitoring data from an environmental monitoring device and acquire original image data of the breeding area from an image acquisition device, wherein the environmental monitoring data includes temperature data, humidity data, and light intensity data; Processing the original image data using an image enhancement algorithm to obtain a preprocessed image data set, wherein the image enhancement algorithm includes an illumination compensation module and an adaptive denoising module; Performing weighted fusion on the environmental monitoring data and the preprocessed image dataset according to the weight parameters output by the duck flock behavior and environment association model to generate a multimodal feature dataset; Determining whether the group behavior indicators in the multimodal feature dataset deviate from a preset threshold range; if so, extracting a spatiotemporal feature graph using a convolutional neural network and combining it with a multi-target tracking algorithm to generate individual behavior trajectory data; Inputting the individual behavior trajectory data into a lightweight neural network model and outputting a health status classification result, wherein the lightweight neural network model is deployed in an edge computing device; generating an environmental control instruction according to the health status classification result, wherein the environmental control instruction includes a temperature compensation amount, a humidity adjustment parameter, and a ventilation intensity index; The environmental control instructions are transmitted to the environmental adjustment device to complete the dynamic environmental optimization of the breeding area.

2. The intelligent monitoring method for duck houses based on image recognition according to claim 1, characterized in that: The steps of acquiring environmental monitoring data from the environmental monitoring device and acquiring original image data of the breeding area from the image acquisition device include: Obtain temperature data, humidity data, and light intensity data from environmental monitoring devices, remove outliers using data cleaning tools, and obtain environmental monitoring data; The original image data of the breeding area is obtained from the image acquisition device.

3. The intelligent monitoring method for duck houses based on image recognition according to claim 1, characterized in that: The step of using an image enhancement algorithm to process the original image data to obtain a preprocessed image data set includes: Acquire original image data of the breeding area from an image acquisition device, process the original image data using a light compensation module, and generate a light-balanced image by adjusting pixel brightness values; For the illumination balanced image, an adaptive denoising module is used to process the illumination balanced image, and noise is removed by a median filtering method to generate a denoised image; Extracting target contour features from the denoised image using an edge detection algorithm, and retaining the target contour features to generate a feature image if the clarity of the target contour is greater than a preset threshold. The characteristic image is used to divide the breeding area in the characteristic image by adopting an image segmentation algorithm, and a monitoring image containing a target area is generated to obtain a preprocessed image data set.

4. The intelligent monitoring method for duck houses based on image recognition according to claim 1, characterized in that: The step of performing weighted fusion on the environmental monitoring data and the pre-processed image dataset based on the weight parameters output by the duck flock behavior and environment association model to generate a multimodal feature dataset includes: Acquire environmental data of temperature, humidity, and ammonia concentration in the breeding area from environmental sensors, process the environmental data using a time series analysis method, and calculate the time series characteristics of each environmental factor using a sliding window method to obtain an environmental characteristic data set; According to the environmental feature dataset and the preprocessed image dataset, weighted fusion is performed on the environmental feature dataset and the preprocessed image dataset using a weight parameter, and an initial multimodal feature set is generated by a matrix weighted summation method to obtain a fused feature set; For the fused feature set, a principal component analysis algorithm is used to perform dimensionality reduction processing on the fused feature set. If the feature dimension after dimensionality reduction is less than a preset threshold, the dimensionality reduction feature is retained to obtain an optimized feature data set; The optimized feature data set is classified and processed by a cluster analysis method, and the monitoring area is divided according to the spatial distribution characteristics of the duck flock behavior to obtain a multimodal feature data set.

5. The intelligent monitoring method for duck houses based on image recognition according to claim 1, characterized in that: The steps of determining whether the group behavior indicators in the multimodal feature data set deviate from a preset threshold range, and if the group behavior indicators in the multimodal feature data set deviate from the preset threshold range, extracting a spatiotemporal feature graph through a convolutional neural network and combining it with a multi-target tracking algorithm to generate individual behavior trajectory data include: Obtaining a group behavior indicator from a multimodal feature data set, and determining whether the indicator deviates from the threshold range by comparing the group behavior indicator with a preset threshold range, thereby obtaining a deviation state; If the deviation state is true, a convolutional neural network is used to process the time series data in the multimodal feature dataset to generate a spatiotemporal feature map; According to the spatiotemporal feature graph, a multi-target tracking algorithm is used to track individuals in the group behavior to generate individual behavior trajectory data.

6. The intelligent monitoring method for duck houses based on image recognition according to claim 1, characterized in that: Inputting the individual behavior trajectory data into the lightweight neural network model and outputting a health status classification result, the steps of deploying the lightweight neural network model in the edge computing device include: Obtaining time series data from an individual behavior trajectory dataset, and performing denoising on the time series data using a filtering algorithm to obtain smooth trajectory data; If the time span of the smooth trajectory data exceeds a preset threshold, extracting the frequency features of the smooth trajectory data by Fourier transform to obtain a frequency feature set; According to the frequency feature set, a lightweight neural network is used to classify the frequency feature set to obtain health status classification data; The health status classification data is transmitted to the dynamic monitoring module through the message queue in the edge computing device to determine whether the health status is abnormal.

7. The intelligent monitoring method for duck houses based on image recognition according to claim 1, characterized in that: The steps of generating an environmental control instruction according to the health status classification result, wherein the environmental control instruction includes a temperature compensation amount, a humidity adjustment parameter, and a ventilation intensity index include: Obtain an abnormal state identifier from the health state classification result, and use a preset mapping table to generate instructions including temperature compensation, humidity adjustment parameters, and ventilation intensity indicators to obtain an initial control instruction set; According to the initial control instruction set, combined with the real-time sensor data in the edge computing device, a weighted average algorithm is used to adjust the temperature compensation amount, humidity adjustment parameter and ventilation intensity index to obtain an optimized control instruction set; If the temperature compensation amount, humidity adjustment parameter or ventilation intensity index in the optimization control instruction set exceeds a preset threshold, the exceeding threshold parameters are calibrated by a fuzzy logic algorithm to obtain a calibration control instruction set; The calibration control instruction set is transmitted to the environmental control module through the message queue in the edge computing device to determine the environmental control instruction.

8. The intelligent monitoring method for duck houses based on image recognition according to claim 1, characterized in that: The steps of transmitting the environmental control instructions to the environmental adjustment device to complete the dynamic environmental optimization of the breeding area include: Obtain the environment control instruction from the edge computing device, transmit it to the environment adjustment device through the message queue protocol, and determine that the transmission of the environment control instruction is completed; According to the temperature compensation amount, humidity adjustment parameter and ventilation intensity index in the environmental control instruction, a preset mapping table is used to adjust the operating state of the environmental control device to obtain an environmental control result; The real-time environmental data of the breeding area is obtained through the built-in sensor of the environmental adjustment device, and the real-time environmental data is compared with the control parameters using a weighted average algorithm to determine whether the dynamic optimization meets the standards; If the dynamic optimization fails to meet the requirements, the control parameters are recalibrated using a fuzzy logic algorithm to obtain an updated control instruction set and transmit it to the environmental control device to determine that the optimization control is completed.

9. An intelligent monitoring system for duck houses based on image recognition, used to implement the intelligent monitoring method for duck houses based on image recognition as described in any one of claims 1 to 8, characterized in that: The intelligent duck house monitoring system based on image recognition includes: A first acquisition module is used to acquire environmental monitoring data from an environmental monitoring device and to acquire original image data of a breeding area from an image acquisition device, wherein the environmental monitoring data includes temperature data, humidity data, and light intensity data; A second acquisition module is used to process the original image data using an image enhancement algorithm to obtain a preprocessed image data set, wherein the image enhancement algorithm includes an illumination compensation module and an adaptive denoising module; A first generation module is configured to perform weighted fusion of the environmental monitoring data and the preprocessed image dataset according to weight parameters output by the duck flock behavior and environment association model to generate a multimodal feature dataset; A second generation module is configured to determine whether the group behavior indicators in the multimodal feature dataset deviate from a preset threshold range. If the group behavior indicators in the multimodal feature dataset deviate from the preset threshold range, a convolutional neural network is used to extract a spatiotemporal feature graph and a multi-target tracking algorithm is used to generate individual behavior trajectory data. an output module, configured to input the individual behavior trajectory data into the lightweight neural network model and output a health status classification result, wherein the lightweight neural network model is deployed in an edge computing device; a third generating module, configured to generate an environmental control instruction according to the health status classification result, wherein the environmental control instruction includes a temperature compensation amount, a humidity adjustment parameter, and a ventilation intensity index; The optimization module transmits the environmental control instructions to the environmental adjustment device to complete the dynamic environmental optimization of the breeding area.

10. The intelligent duck house monitoring system based on image recognition according to claim 9, characterized in that: The first acquisition module includes: A first acquisition unit is configured to acquire temperature data, humidity data, and light intensity data from an environmental monitoring device, and remove abnormal values using a data cleaning tool to obtain environmental monitoring data; The second acquisition unit is used to acquire original image data of the breeding area from the image acquisition device.

Images