A comprehensive detection method for intestinal development and function of weaned piglets

Abstract

The present application relates to the field of livestock breeding and intelligent detection technology, and specifically discloses a comprehensive detection method for weaned piglet intestinal development and function, comprising: collecting weaned piglet feces and abdominal wall near-infrared spectrum, fusing DNA methylation markers and near-infrared fluorescent nanosensor, and non-invasively positioning intestinal micro-injury area; collecting weaned piglet intestinal contents at different ages, constructing a metabolic group time sequence atlas, fusing an adaptive neural fuzzy reasoning system, backtracking the acute or chronic course of the positioned injury, and distinguishing between stress injury and development lag; the present application generates a three-dimensional intestinal injury probability heat map without collecting intestinal biopsy tissue by fusing fecal exfoliative cell DNA methylation detection and abdominal wall transmission spectrum analysis of near-infrared fluorescent nanosensor, can detect early hidden injuries that are difficult to find by conventional means, and can distinguish between primary injury and secondary spread area, thereby significantly reducing the missed detection rate of weaned piglet intestinal lesions.

Description

Technical Field

[0001] This invention relates to the fields of animal husbandry and intelligent detection technology, and in particular to a comprehensive detection method for the intestinal development and function of weaned piglets. Background Technology

[0002] The weaning stage is a critical period for piglet growth and development. During this time, piglets transition from breast milk to solid feed, and the adaptability and function of their intestines have a significant impact on their growth, health, and production performance. Piglet intestinal development involves multiple aspects, including cell proliferation, differentiation, and the formation of the microbial community, affecting their digestive and absorptive capacity and the establishment of their immune system.

[0003] For example, Chinese Patent Publication No. CN115308424A describes a method for detecting weaning stress-induced immune responses in piglets based on biomass spectrometry. This method effectively eliminates matrix effects caused by randomness due to sample differences and fluctuations in complex ion signals, reduces sensitivity to ionization parameters, and minimizes the loss of ion peak information, thereby improving the detection accuracy of ion peaks.

[0004] In existing detection techniques, when faced with occult intestinal mucosal micro-injuries induced by weaning stress, routine biopsies of weaned piglets are difficult to non-invasively locate the damaged area, easily leading to missed detection of early lesions. Moreover, detection data at a single time point is not convenient to distinguish whether the damage originates from acute stress or chronic developmental delay, and lacks the ability to trace the temporal evolution of the damage. Furthermore, existing methods are difficult to predict whether the current intestinal microenvironment supports epithelial function reconstruction, making interventions such as supplemental feeding lack specificity. Therefore, how to achieve a progressive and comprehensive detection from non-invasive localization of occult damage, to temporal tracing of damage evolution, and then to functional adaptation prediction is the problem that this invention aims to solve. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, the present invention provides a comprehensive detection method for the intestinal development and function of weaned piglets, which can effectively solve the problems involved in the prior art.

[0006] The objective of this invention can be achieved through the following technical solution: This invention provides a comprehensive detection method for the intestinal development and function of weaned piglets, comprising the following steps:

[0007] Step 1: Collect near-infrared spectra of feces and abdominal wall of weaned piglets, and integrate DNA methylation markers and near-infrared fluorescent nanosensors to non-invasively locate intestinal micro-damage areas and achieve accurate detection of early occult damage.

[0008] Step 2: Collect intestinal contents of weaned piglets at different ages, construct metabolome time series map, integrate adaptive neurofuzzy reasoning system, retrospectively analyze the acute or chronic course of the located injury, distinguish between stress injury and developmental delay, have the ability to trace the temporal evolution of injury, and accurately distinguish between acute stress and chronic developmental delay.

[0009] Step 3: Overlay deep Q-network with real-time streaming data from wearable multi-parameter biochemical sensors to dynamically extrapolate the intestinal epithelial repair threshold, predict whether the current intestinal microenvironment supports epithelial function reconstruction, achieve quantitative prediction of intestinal epithelial repair conditions, and provide a scientific basis for intervention decisions;

[0010] Step 4: Based on the estimated repair threshold, assess whether the current gut microenvironment supports epithelial reconstruction, predict the adaptive recovery potential of piglet gut development and function, accurately assess the microenvironment's support for epithelial reconstruction, and improve the accuracy of recovery potential prediction.

[0011] Step 5: Based on the prediction results, adaptively trigger supplementary feeding, enzyme preparation or microbial intervention instructions to achieve individualized regulation of intestinal development and function detection in weaned piglets and improve the efficiency of intestinal function recovery in weaned piglets.

[0012] Step 6: Integrate location, time series, prediction and intervention data to generate a report on the intestinal development and function of weaned piglets, realizing a full-chain assessment from the identification of insidious damage to the prediction of functional adaptation, and supporting precision breeding management.

[0013] Preferably, step 1 specifically includes:

[0014] Fresh fecal samples were collected from weaned piglets. The methylation level of CpG islands in the promoter regions of the tight junction proteins Occludin and Claudin-1 in the fecal exfoliated intestinal epithelial cells was extracted using a magnetic bead capture method. The results were combined with a near-infrared fluorescent nanosensor (surface-modified with antibodies targeting myeloperoxidase and tight junction protein-1) excited by transmission through the abdominal wall after mixing with feces. The spatial distribution signals of intestinal permeability and inflammatory activity were simultaneously acquired to achieve non-invasive early molecular recognition of insidious damage.

[0015] The acquired quantitative methylation data is spatiotemporally registered with the near-infrared spectral characteristic peak intensities corresponding to different intestinal segments (jejunum, ileum, and colon). A methylation-spectral joint feature map is constructed using a convolutional neural network to achieve spatial coupling analysis of methylation abnormalities at a single site and large-area spectral signal attenuation regions, thereby generating pixel-level intestinal damage probability maps to achieve precise spatial localization and quantitative assessment of the damaged area.

[0016] Based on the generated intestinal injury probability map, an adaptive confidence threshold (dynamically adjusted with the age of piglets and the number of days since weaning) is set to balance the positive predictive value and the false positive rate. The spatial coordinates and range of high-probability cryptic micro-injury areas are automatically segmented and labeled from the map, and a three-dimensional injury heat map with intestinal anatomical location as a reference is output, realizing the visualization output of the injury area and the accurate labeling of the anatomical location.

[0017] Preferably, step 1 further includes:

[0018] Near-infrared spectral time-series data of the abdominal wall of the same piglet were collected on days 0, 3, 7, 10 and 14 after weaning. Fluorescence intensity change curves of projection points on each body surface were constructed. The spatial migration trajectory of the near-infrared fluorescent nanosensor signal was dynamically tracked to realize dynamic visualization tracking of the spatial location of the damaged area over time.

[0019] The methylation detection results were stratified by time point, and the consistency between the spatial distribution of DNA methylation spectrum and fluorescence signal at different ages was compared using the dynamic time warping algorithm. This helped to identify whether the damaged area expanded or migrated, and to quantitatively assess the spatiotemporal evolution and consistency of the damaged area.

[0020] Based on the direction of signal migration and the rate of change in methylation, the initial site of insidious damage and the secondary diffusion path can be located, the spatial range of primary and secondary damage can be distinguished, and the primary and secondary diffusion regions can be accurately delineated.

[0021] Preferably, step 2 specifically includes:

[0022] The contents of the terminal ileum of piglets were collected on days 0, 2, 5, 9, 14 and 21 after weaning. The relative abundance of small molecule metabolites such as short-chain fatty acids, tryptophan metabolites and secondary bile acids was obtained by ultra-high performance liquid chromatography-tandem mass spectrometry. The number of exfoliated cells in the intestinal mucosa in each sample was recorded simultaneously as a tissue background correction parameter for the source of metabolites, eliminating the interference of exfoliated cells and improving the accuracy of metabolite quantification.

[0023] We used sparse partial least squares discriminant analysis based on elastic network regularization to reduce the dimensionality of time-series metabolic data. We automatically screened out three types of characteristic metabolite combinations that showed monotonically increasing, monotonically decreasing, or U-shaped changes with the weaning age. We constructed a metabolomics time-series map with age as the horizontal axis, screened dynamic characteristic metabolites, and revealed the evolution of the intestinal barrier.

[0024] The rate of change of each characteristic metabolite at adjacent time points was calculated. Combined with the dynamic change curve of intestinal mucosal exfoliated cell count, the inflection points and fluctuation ranges in the metabolic trajectory that exceed the 95% confidence interval of healthy piglets of the same age were marked. A metabolomic time-series fingerprint database reflecting the dynamic evolution of intestinal mucosal barrier function was formed, the fluctuation of time-series inflection points was quantified, and the barrier function fingerprint was constructed.

[0025] Preferably, step 2 further includes:

[0026] The time-series curves of each metabolite in the metabolome time-series fingerprint database are input into an adaptive neurofuzzy reasoning system that integrates attention mechanism. The time point is used as the antecedent input and the damage course type is used as the consequent output. Prior knowledge of the causal relationship between metabolites is introduced as the initial weight of the fuzzy rules. Three fuzzy membership functions are set to improve the physiological mechanism fit of the disease course type discrimination.

[0027] The pseudo-labels generated by time-series hierarchical clustering were used for semi-supervised pre-training to initialize the fuzzy rule base. The rule parameters were adjusted by a hybrid learning algorithm that alternately optimized backpropagation and least squares estimation, so as to maximize the weighted Kappa coefficient of the system output discrimination result and the actual intestinal mucosal histological score of the piglet on the 28th day, thereby enhancing the consistency between the model and the histological gold standard.

[0028] The system outputs an acute stress index and a developmental lag index after training. The ratio of these two indices is used as a comprehensive disease course discrimination factor. If the factor is greater than 2.0 and U-shaped metabolites dominate in the metabolic trajectory, it is judged as acute stress injury. If the factor is less than 0.5 and monotonically decreasing metabolites dominate, it is judged as chronic developmental lag. The boundary interval is judged as mixed injury, which clearly quantifies the disease course type and provides traceable evidence.

[0029] Preferably, step 3 specifically includes:

[0030] Wearable multi-parameter biochemical sensors were fixed to the subcutaneous tissue behind the ears of piglets. Continuous streaming data of intestinal fatty acid binding protein, pH value, lactic acid and glucose concentration were collected using a variable interval sampling strategy. Kalman filtering was used to remove motion artifacts, achieving non-invasive continuous monitoring, effectively filtering out motion interference, and obtaining stable and reliable time-series data of physiological parameters.

[0031] An environmental state vector is constructed by combining current sensor readings with the obtained damage location and disease course information. The action space of the deep Q network is defined to include three types of repair actions: promoting proliferation, anti-inflammatory regulation, and resting observation. Multi-source information is integrated to construct a state representation, providing a complete environmental perception basis for intelligent decision-making.

[0032] By setting the in vitro migration rate of crypt cells and the organoid formation rate of intestinal epithelium as reward signals, the value of actions is explored through an ε-greedy strategy. The weights of the Q network are iteratively updated until the state-action value function converges, driving the model to autonomously optimize the intervention strategy and achieve learning convergence from data to the optimal repair decision.

[0033] Preferably, step 3 further includes:

[0034] The optimal action value function is extracted from the convergent deep Q-network model, and the maximum threshold of intestinal epithelial proliferation rate corresponding to the execution of repair actions under various intestinal microenvironment states is deduced. A repair threshold surface is constructed to realize the quantitative mapping between repair conditions and action effects, thereby improving the scientific nature of threshold setting.

[0035] The current real-time readings of wearable sensors are mapped to the corresponding coordinate points of the threshold surface. The Euclidean distance of each continuity parameter from the surface boundary is calculated. The minimum energy substrate concentration and maximum inflammatory tolerance level required to maintain epithelial reconstruction are dynamically quantified, realizing real-time quantitative assessment of the microenvironment state and providing dynamic basis for intervention decisions.

[0036] The system outputs a three-tiered repair threshold range based on short-chain fatty acid, endotoxin concentration, and redox potential, corresponding to three microenvironmental levels: high risk, critically repairable, and supportive of repair. This provides a clear classification of the microenvironmental repair level and helps to quickly determine the urgency of intervention.

[0037] Preferably, step 4 specifically includes:

[0038] The system collects real-time values ​​of three key parameters from wearable sensors, representing energy metabolism (lactic acid and glucose), barrier integrity (intestinal fatty acid-binding proteins), and inflammatory status (real-time indirectly estimated pH fluctuations), enabling multimodal synchronous capture of key microenvironment indicators.

[0039] The real-time values ​​of the three key parameters are compared with the corresponding dimension repair thresholds. If all three parameters are simultaneously better than or reach the support repair range, the high recovery potential is determined. If any one of them is lower than the critical repairable lower limit, the low recovery potential is determined, thus realizing the real-time classification and determination of the microenvironment support capability.

[0040] For the determination of low recovery potential, the parameter dimension that first breaks through the lower limit is further identified as the bottleneck factor, and the bottleneck factor type label and the quantitative difference of deviation from the threshold are output to achieve precise positioning of the key obstacles restricting the repair.

[0041] Preferably, step 5 specifically includes:

[0042] An intervention strategy library is constructed, in which energy bottlenecks are matched with supplemental feeding instructions (quantitative addition of lactose or medium-chain triglycerides), digestion bottlenecks are matched with enzyme preparation instructions (combination ratio of protease and amylase), and microbial bottlenecks are matched with microbial intervention instructions (combination of lactic acid bacteria and Bacillus). For each bottleneck factor, two or more compensatory paths and switching priorities are preset. When the preferred path fails due to coexistence constraints, it is automatically downgraded to the secondary path to ensure that there are targeted intervention plans for different bottlenecks. The automatic path switching ensures the reliability of execution.

[0043] The system receives the bottleneck factor type label and the quantitative difference of the deviation threshold, calls the matching instruction and calculates the initial intervention dose based on the deviation threshold difference, and determines the step-reduction dosing regimen for three consecutive days by looking up a table, so as to achieve precise matching between the dose and the degree of damage, and the step-reduction avoids over-intervention and promotes smooth recovery.

[0044] The final generated intervention instructions are encoded into standard aquaculture management messages, pushed to the automated feeding system via wireless network, and locked to prevent repeated triggering within the execution cycle. At the same time, the instruction triggering time and parameters are recorded for subsequent reporting, realizing automated closed-loop transmission and prevention of repeated execution of intervention instructions, and supporting full-process traceability and effect evaluation.

[0045] Preferably, step 6 specifically includes:

[0046] By integrating the output damage heatmap time-series snapshots, metabolic trajectories and disease course classification labels, repair threshold ranges and potential grading and intervention records, a multi-source heterogeneous dataset is constructed to achieve unified management and rapid retrieval of multi-dimensional detection data;

[0047] The execution report is templated and filled in, with a color spatial distribution map to present the damage heat map, a metabolic trajectory to be presented as an age-abundance line graph overlaid with a fuzzy system discrimination mark, and a repair threshold range to be displayed in three dimensions in the form of a three-dimensional surface, intuitively presenting the spatial relationship between damage evolution, metabolic trajectory and repair conditions.

[0048] The report generates a conclusion paragraph, arranged in the order of location results, time-series attribution, current repair conditions, intervention recommendations, and reassessment time points. The reassessment time points are dynamically predicted by the deep Q network based on the current repair speed rather than at a fixed period. All original data is also accompanied by tamper-proof hash verification codes and QR code index links, providing a dynamic assessment cycle and ensuring the reliable traceability of data across the entire chain.

[0049] Compared with the prior art, the beneficial effects of the present invention are:

[0050] 1. This comprehensive detection method for intestinal development and function in weaned piglets integrates fecal exfoliated cell DNA methylation detection with abdominal wall transmission spectroscopy analysis using a near-infrared fluorescent nanosensor. It can generate a three-dimensional intestinal injury probability heatmap without collecting intestinal biopsy tissue, and can detect early occult lesions that are difficult to detect by conventional methods. It can also distinguish between primary lesions and secondary spread areas, significantly reducing the missed detection rate of intestinal lesions in weaned piglets.

[0051] 2. This comprehensive detection method for intestinal development and function in weaned piglets constructs time-series maps of the intestinal metabolome at multiple time points after weaning, and combines an adaptive neural fuzzy inference system to perform time-series analysis of metabolic trajectories. It can accurately determine whether the source of the located damage is an acute stress response or chronic developmental delay, making up for the technical deficiency of single-time-point detection in being unable to trace the disease progression, and providing etiological basis for targeted intervention.

[0052] 3. This comprehensive detection method for intestinal development and function in weaned piglets uses a deep reinforcement learning network to fuse and analyze real-time monitoring data from wearable sensors with damage status information, dynamically infers the repair threshold required for intestinal epithelial function reconstruction, and outputs a graded microenvironment support level. This overcomes the limitation of existing technologies that cannot predict whether the intestinal microenvironment has the ability to repair itself, and provides a quantitative basis for intervention decisions.

[0053] 4. This comprehensive detection method for intestinal development and function in weaned piglets automatically matches supplementary feeding, enzyme preparations, or microbial programs from the intervention strategy library based on the bottleneck factor identification results and the deviation threshold quantification difference, and generates a step-by-step dose reduction instruction to push to the automated feeding system, realizing closed-loop regulation from detection to intervention, avoiding the blindness of empirical intervention, and improving the efficiency and targeting of intestinal function recovery in weaned piglets. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the workflow of a comprehensive detection method for intestinal development and function in weaned piglets according to the present invention.

[0055] Figure 2 This is a schematic diagram of the process flow for a comprehensive detection method for intestinal development and function in weaned piglets according to the present invention. Detailed Implementation

[0056] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0057] Example 1, please refer to Figure 1 , Figure 2 This invention provides a technical solution: a comprehensive detection method for the intestinal development and function of weaned piglets, comprising the following steps:

[0058] Step 1: Near-infrared spectra of feces and abdominal wall from weaned piglets are collected. DNA methylation markers and near-infrared fluorescent nanosensors are fused to non-invasively locate intestinal micro-damage areas, enabling accurate detection of early, insidious injuries and avoiding missed lesions. Fresh fecal samples naturally excreted from weaned piglets are collected. The methylation levels of CpG islands in the promoter regions of the tight junction proteins Occludin and Claudin-1 in exfoliated intestinal epithelial cells are extracted using a magnetic bead capture method. This data is then combined with near-infrared fluorescent nanosensors (surface-modified with antibodies targeting myeloperoxidase and tight junction protein-1) excited by transmission through the abdominal wall after mixing with feces. Simultaneously, spatial distribution signals of intestinal permeability and inflammatory activity are acquired, enabling non-invasive early molecular recognition of insidious injuries. The acquired quantitative methylation data is spatiotemporally registered with the near-infrared spectral characteristic peak intensities corresponding to different intestinal segments (jejunum, ileum, and colon). A methylation-spectral joint feature map is constructed using a convolutional neural network to achieve spatial coupling analysis of methylation abnormalities at a single site and large-area spectral signal attenuation regions. This generates a pixel-level intestinal injury probability map, achieving precise spatial localization and quantitative assessment of the damaged area. Based on the generated intestinal injury probability map, an adaptive confidence threshold (dynamically adjusted with piglet age and weaning days) is set to balance positive predictive values ​​and false positive rates. The spatial coordinates and ranges of high-probability cryptic micro-damage areas are automatically segmented and labeled from the map, and a three-dimensional damage heat map with intestinal anatomical location as a reference is output, realizing the visualization output of the damaged area and precise anatomical location calibration.

[0059] It should be noted that fresh fecal samples were collected from weaned piglets, with a sample volume of 2.0 ± 0.2 g placed in a sterile centrifuge tube. 5 mL of pre-chilled phosphate buffer (4°C) was added, and the mixture was vortexed for 30 seconds before centrifugation to remove the supernatant. This process was repeated twice. Magnetic bead capture lysis buffer (containing 50 mM Tris-HCl, 10 mM EDTA, and 0.5% SDS) was added to the precipitate. (w / v and proteinase K 0.2 mg / mL) was lysed in a 56°C water bath for 120 min. Biotin-labeled methylation probes were captured using streptavidin-modified magnetic beads. Methylation was quantified targeting the CpG islands in the Occludin gene promoter region (sites 114,000, 100-114,000, 350 on chromosome 5 of the hg38 genome) and the CpG islands in the Claudin-1 gene promoter region (sites 190, 200, 450-190, 200, 700 on chromosome 3). Simultaneously, surface-modified targets for myelin peroxidation were also analyzed. Near-infrared fluorescent nanosensors (particle size 80 nm, excitation wavelength 785 nm, emission wavelength 1100 nm) containing enzyme antibodies and tight junction protein-1 antibodies were administered orally at a dose of 0.5 mg / kg body weight, mixed with feces. After 180 minutes, an 850 nm laser was used to transmit light through the abdominal wall, and the emission spectrum was collected using an InGaAs array detector. Fluorescence intensity values ​​were obtained from eight surface projection points in each of the three segments: jejunum, ileum, and colon. Areas with fluorescence attenuation exceeding 40% of the baseline value were identified as areas of inflammation or abnormal permeability. Methylation quantification results were expressed as percentages (0.5 mg / kg body weight, 785 nm excitation wavelength, 1100 nm emission wavelength). (-100%) indicates that the near-infrared spectral characteristic peak intensities (normalized relative intensity values) of the corresponding surface projection points of each intestinal segment are spatiotemporally registered according to anatomical coordinates. A three-layer convolutional neural network is constructed: the input layer receives a methylation-spectral dual-channel data matrix (64×64 pixels resolution), the first convolutional layer uses a 3×3 convolutional kernel (32 channels, stride 1), and the pooling layer uses 2×2 max pooling; the second convolutional layer uses a 3×3 convolutional kernel (64 channels), and the third convolutional layer uses a 3×3 convolutional kernel (128 channels); the fully connected layer outputs a 32-dimensional feature vector. Damage probabilities were mapped to the 0-1 interval using the Sigmoid activation function. The network was trained using a transfer learning strategy, with pre-training weights derived from a dataset of 200 piglet intestinal samples annotated with histopathology (sensitivity 0.92, specificity 0.88). After outputting pixel-level intestinal damage probability maps, adaptive confidence thresholds were set: 0.62 for days 0-3 post-weaning, 0.68 for days 4-7, 0.72 for days 8-14, and 0.69 for days 15-21. These thresholds were dynamically adjusted to maintain a positive predictive value ≥0.85 and a false positive rate ≤0.12. Regions with a probability ≥ a threshold and a connected area ≥ 5 pixels were segmented using a region growing algorithm as lesion candidate regions. Morphological processing was performed on the segmented lesion candidate regions, sequentially executing a 3×3 structuring element erosion operation to eliminate isolated noise points and a 5×5 structuring element dilation operation to fill internal voids. Finally, the centroid coordinates, circumscribed rectangle boundaries, and pixel areas of each connected region were extracted. Based on a pre-established spatial mapping model of abdominal wall surface projection points and intestinal anatomical locations (this model was constructed based on CT image registration data of 30 piglets, with coordinates...),... (Conversion error ≤ 2.5mm) The coordinates of the damaged area in the image coordinate system are transformed to a three-dimensional intestinal anatomical coordinate system with the anus as the origin and the jejunum to the colon as the vertical axis. The annotations include: the distribution percentage of the damaged area in each segment of the jejunum, ileum, and colon (as a percentage of the total pixel area of ​​that segment), the centroid coordinates and maximum diameter of the largest damaged cluster, and finally, a three-dimensional damage heatmap is output, using Jet color scales to map the damage probability values ​​(0.62-0.80 is mapped to yellow to orange, and 0.80-1.00 is mapped to red).

[0060] Furthermore, step 1 also includes: collecting near-infrared spectral time-series data of the abdominal wall of the same piglet on days 0, 3, 7, 10, and 14 after weaning, constructing fluorescence intensity change curves of projection points on each body surface, dynamically tracking the spatial migration trajectory of near-infrared fluorescent nanosensor signals, realizing dynamic visualization tracking of the spatial location of the damaged area over time, stratifying the methylation detection results by time point, using a dynamic time warping algorithm to compare the consistency between the spatial distribution of DNA methylation spectra and fluorescence signals at different ages, identifying whether the damaged area has expanded or migrated, quantitatively assessing the spatiotemporal evolution and consistency of the damaged area, locating the initial occurrence site and secondary diffusion path of the insidious damage based on the signal migration direction and the rate of change of methylation degree, distinguishing the spatial range of primary and secondary damage, and accurately dividing the primary damage and secondary diffusion areas;

[0061] It should be noted that near-infrared spectral data of the abdominal wall of the same weaned piglet were collected at fixed time points on days 0, 3, 7, 10, and 14 after weaning (9:00-10:00 AM daily, after a 2-hour fast). Before testing, the piglets were restrained. An 850nm laser was used to transmit the signal through the abdominal wall, and an InGaAs array detector was used to sequentially scan eight surface projection points in each of the three segments: jejunum (2-5 cm below the xiphoid process), ileum (periumbilical region), and colon (left lower abdomen and suprapubic region), for a total of 24 points. Spectral signals were collected three times consecutively at each point, and the average value was taken as the mean. To minimize fluctuations caused by respiratory movements and intestinal peristalsis, the fluorescence intensity values ​​at each location were arranged by time point. Using the baseline fluorescence intensity on the day of weaning (Day 0) as a reference, the relative percentage of fluorescence intensity at each time point was calculated. Cubic spline interpolation was used to smooth the discrete time-point data for each location, constructing a fluorescence intensity curve for each projected point over time. Based on the curve trend, the locations were categorized as stable (curve fluctuation amplitude ≤ ±10%), decaying (decay ≥ 15% at two or more consecutive time points), and fluctuating (amplitude exceeding ±20%). The results were categorized into three types: (1) with no monotonic trend, and (2) with attenuation-type sites indicating a continuous decrease in nanosensor signals, reflecting an increase in permeability or inflammation in the corresponding intestinal segment over time. The methylation results of the Occludin and Claudin-1 gene promoters were stratified according to days 0, 3, 7, 10, and 14. The methylation level of each intestinal segment corresponding to the surface projection point at each time point was expressed as a percentage. Simultaneously, the fluorescence intensity values ​​of the 24 surface projection points at each time point were normalized to construct a 24-dimensional fluorescence signal spatial distribution vector. Dynamic analysis was then performed. The temporal warping algorithm compares the consistency of methylation spectra (12-dimensional, containing 6 representative sites from each of the two genes intestinal segments) with the spatial distribution vectors of fluorescence signals at different ages. Specifically, a Sakoe-Chiba window is set as the local constraint, with the window width limited to 10% of the time axis length. Euclidean distance is used as the distance metric, calculating the warped path distance for each time point combination. A warped distance less than 0.3 is considered highly consistent, 0.3 to 0.6 is moderately consistent, and greater than 0.6 is inconsistent. A warped distance exceeding 0.3 is considered inconsistent when two or more consecutive time points show an increase in warped distance exceeding 0.At 2 o'clock, it was determined that the lesion area had expanded or migrated. Simultaneously, the deviation direction of the regular path at adjacent time points was compared. If the regular path deviated towards the colon and the fluorescence attenuation point simultaneously shifted downwards towards the abdominal wall, it was determined that the lesion had spread from the jejunum or ileum towards the colon. Based on the signal migration direction and the rate of change in methylation level, the initial site of the occult lesion and the secondary diffusion path were located. The point with the most significant fluorescence attenuation (relative fluorescence intensity ≤ 60% of baseline) and the highest methylation level on day 0 after weaning was identified as the candidate site for primary lesion. The rate of change in fluorescence intensity and the rate of change in methylation at this candidate site were tracked time-by-time (difference in methylation percentage between adjacent time points divided by the number of days). If the fluorescence intensity continued to decrease and the methylation rate was ≥ 8% / day, it was confirmed as the primary lesion. For primary lesion sites, for newly appearing fluorescence decay sites at subsequent time points (decay exceeding 40% and previously normal), the cross-correlation function between their fluorescence intensity change curves and the primary lesion site change curves is calculated. If the cross-correlation coefficient is ≥0.75 and the time delay is within 1-3 days, it is determined to be a secondary lesion site. A diffusion path is generated by connecting the centroid coordinates of all secondary sites, and the diffusion rate (pixel area increase divided by the time interval) is calculated. Finally, the lesion area is divided into a primary region (containing the initial site and a continuous area ≥10 pixels) and a secondary region (distributed along the diffusion path and not directly connected to the primary region but with temporal correlation). The intestinal segment distribution ratio, maximum diameter length, and average lesion probability of the two types of regions are output respectively.

[0062] Step 2: Intestinal contents were collected from weaned piglets at different ages to construct a metabolomics time-series map. An adaptive neurofuzzy reasoning system was integrated to retrospectively analyze the acute or chronic course of localized injuries, distinguishing between stress-induced injury and developmental delay. The system possesses the ability to trace the temporal evolution of injuries and accurately differentiate between acute stress and chronic developmental delay. Ileal contents were collected from piglets on days 0, 2, 5, 9, 14, and 21 post-weaning. Ultra-high performance liquid chromatography-tandem mass spectrometry (UHPLC-MS / MS) was used to obtain the relative abundance of small molecule metabolites such as short-chain fatty acids, tryptophan metabolites, and secondary bile acids. Simultaneously, the number of exfoliated intestinal mucosal cells in each sample was recorded as a tissue background correction parameter to eliminate interference from exfoliated cells. To improve the accuracy of metabolite quantification, sparse partial least squares discriminant analysis based on elastic network regularization was used to reduce the dimensionality of time-series metabolic data. Three types of characteristic metabolite combinations with monotonically increasing, monotonically decreasing, or U-shaped changes with weaning age were automatically screened. A metabolomics time-series map with age as the x-axis was constructed to screen dynamic characteristic metabolites, reveal the evolution law of the intestinal barrier, calculate the rate of change of each characteristic metabolite at adjacent time points, and combine the dynamic change curve of intestinal mucosal exfoliated cell count to mark the inflection points and fluctuation ranges in the metabolic trajectory that exceed the 95% confidence interval of healthy piglets of the same age. This formed a metabolomics time-series fingerprint library reflecting the dynamic evolution of intestinal mucosal barrier function, quantified the fluctuation of time-series inflection points, and constructed a barrier function fingerprint.

[0063] It should be noted that on days 0, 2, 5, 9, 14, and 21 post-weaning, between 9:00 and 10:00 AM (after a 2-hour fast), the same weaned piglets were restrained, and ileal contents were collected using a sterile rectal swab or a fistula in the terminal ileum. The sample size was 1.0 ± 0.1 g, immediately frozen in liquid nitrogen, and thawed at 4°C before testing. Protein precipitation was performed by adding a pre-cooled acetonitrile-water mixture (1:1, v / v), followed by ultrasonic extraction for 10 minutes and centrifugation (12000 rpm, 10 minutes, 4°C). The supernatant was filtered through a 0.22 μm filter membrane, and metabolite detection was performed using ultra-high performance liquid chromatography-tandem mass spectrometry (UPLC-MS / MS). The chromatographic column was an ACQUITY UPLC HSS. T3 (2.1×100mm, 1.8μm), mobile phase A was 0.1% formic acid aqueous solution, mobile phase B was 0.1% formic acid acetonitrile solution, gradient elution program was 0-2 min 5%B, 2-12 min 5%-95%B, 12-14 min 95%B, 14-14.5 min 95%-5%B, 14.5-15 min 5%B, flow rate 0.3 mL / min, column temperature 40℃, mass spectrometry used electrospray ionization (ESI) negative ion mode, ion source temperature 350℃, spray voltage -3.0 kV, scan range m / z The relative abundance (expressed as peak area, normalized by internal standard) of small molecule metabolites such as short-chain fatty acids (acetic acid, propionic acid, butyric acid, isobutyric acid, valeric acid), tryptophan metabolites (kynurenine, 5-hydroxytryptamine, indole-3-acetic acid, tryptophan), and secondary bile acids (deoxycholic acid, lithocholic acid, ursodeoxycholic acid) in samples was obtained from 50 to 1200. Simultaneously, the number of exfoliated intestinal mucosal cells in each sample was recorded: 10 μL of the contents suspension was placed in a hemocytometer, and the number of epithelial cells was counted under a light microscope as a tissue background correction parameter for metabolite origin. The correction formula is Corrected abundance = Original abundance / (Exfoliated cell count + 1) to reduce the bias caused by differences in intestinal mucosal shedding on metabolite quantification. The corrected relative abundance data of metabolites at six time points are used to construct a time-metabolite matrix. Sparse partial least squares discriminant analysis based on elastic net regularization is used for dimensionality reduction. The elastic net mixing parameter α is set to 0.5, and the regularization parameter λ is selected to minimize the classification error rate through 10-fold cross-validation. Weaning age is used as the classification variable, and cells with absolute loading values ​​greater than 0 are automatically selected.Metabolites of metabolites 3 were further categorized into three types based on their abundance trends with age: monotonically increasing (positive rates of change at adjacent time points and an overall slope > 0), monotonically decreasing (negative rates of change at adjacent time points and an overall slope < 0), and U-shaped (decreasing first and then increasing, with the inflection point occurring between days 5 and 9, and the trough abundance being less than 80% of the endpoint abundance). A metabolomics time series map was constructed with age as the x-axis and the normalized abundance (mean ± standard deviation) of various metabolites as the y-axis. The time-series curves for each metabolite were smoothed using a three-point moving average. The graphs also included reference ranges for age-matched healthy piglets (95% confidence intervals constructed based on historical data from 50 normally weaned piglets; age-specific confidence bands were calculated for each metabolite). This graph visually demonstrates the dynamic evolution of candidate metabolites throughout the weaning process. The rate of change for each characteristic metabolite at adjacent time points (days 0-2, 2-5, 5-9, 9-14, and 14-21) was calculated one by one. , For the rate of change, For time points abundance, For time points Abundance and rate of change are expressed as a percentage change in daily abundance. Combined with the dynamic curve of intestinal mucosal exfoliated cell count, the time interval of abnormally high exfoliated cell count (exceeding the baseline mean + 2 standard deviations, with the baseline being the health status value on day 0) is marked as a high exfoliation period. For each metabolite, the region in its metabolic trajectory that exceeds the 95% confidence interval of healthy piglets of the same age is marked, and inflection points are further identified: the time point where the rate of change changes from positive to negative or from negative to positive and the absolute value of the rate change before and after the inflection point is greater than twice the average rate of the two adjacent points. The fluctuation interval is defined as the region where two or more consecutive time points exceed the confidence interval and the direction of the rate of change is inconsistent (containing at least one peak and one trough). All the time-series metabolic data marked above are summarized to form a metabolomics time-series fingerprint database reflecting the dynamic evolution of intestinal mucosal barrier function. Each fingerprint record includes: metabolite name, category label (monotonically increasing / decreasing / U-shaped), corrected abundance at each time point, rate of change at adjacent time points, start and end time of the exceeding interval, inflection point position (days), and fluctuation interval range.

[0064] Furthermore, step 2 also includes: inputting the time-series curves of each metabolite in the metabolome time-series fingerprint database into an adaptive neurofuzzy inference system that integrates an attention mechanism, using time points as antecedents and damage course type as consequents as outputs, introducing prior knowledge of causal relationships between metabolites as initial weights for fuzzy rules, setting three fuzzy membership functions to improve the physiological mechanism fit for course type discrimination, using pseudo-labels generated by time-series hierarchical clustering for semi-supervised pre-training to initialize the fuzzy rule base, and adjusting the rule parameters through a hybrid learning algorithm that alternates between backpropagation and least squares estimation optimization, so that the system... The weighted Kappa coefficient of the system output discrimination result and the actual intestinal mucosal histological score of the piglet on the 28th day is maximized to enhance the consistency between the model and the histological gold standard. The system outputs the acute stress index and the developmental lag index after training, and the ratio of the two is introduced as a comprehensive disease course discrimination factor. If the factor is greater than 2.0 and the U-shaped characteristic metabolites dominate in the metabolic trajectory, it is judged as acute stress injury. If the factor is less than 0.5 and the monotonically decreasing characteristic metabolites dominate, it is judged as chronic developmental lag. The boundary interval is judged as mixed injury, which clearly quantifies the disease course type and provides traceable evidence.

[0065] It should be noted that six time points after weaning were used as antecedent input variables, and the type of injury course (acute stress type, chronic developmental delay type, and mixed type) was used as consequent output variables. Three membership functions were set, corresponding to low, medium, and high fuzzy levels, respectively. The membership functions adopted the Gaussian type. The initial values ​​of the center parameter were set according to the 25th, 50th, and 75th percentiles of metabolite abundance in healthy piglets. The width parameter was uniformly set to 0.3 times the difference between the 90th and 10th percentiles. Prior knowledge of the causal relationship between metabolites was introduced as the initial weight of the fuzzy rules: the activation of the inflammatory response via the kynurenine pathway by the increase of tryptophan metabolite was set as high weight (0.8), and the association between the decrease of short-chain fatty acids and barrier function damage was set as medium weight (0.6). The association between changes in bile acids and dysbiosis was assigned a low weight (0.4). The output layer used a weighted average method to defuzzify and generate membership vectors for each disease course type. The time-metabolite matrix composed of the corrected abundance of each metabolite at six time points was input into a time-series hierarchical clustering algorithm. Dynamic time-normalized distance was used as a similarity measure. The Sakoe-Chiba window width was set to 20% of the time axis length. The number of clusters was preset to 3, corresponding to the three disease course types. Pseudo-labels for each sample were generated as pre-training supervision signals. The fuzzy rule base was initialized with 27 rules (a combination of three inputs and three membership degrees). The initial confidence of each rule was set to the majority vote ratio of the pseudo-labels in the pre-training samples covered by the rule. Backpropagation and least squares estimation were used alternately. The optimized hybrid learning algorithm adjusts the rule parameters: the antecedent parameters (center and width of the membership function) are updated through backpropagation, with a learning rate of 0.01 and a momentum factor of 0.9; the consequent parameters are solved using a closed-form least-squares estimation. The training objective is to maximize the weighted Kappa coefficient between the system output discrimination result and the actual intestinal mucosal histological score of the piglet on day 28. The histological score uses a four-level scale of 0-3, and the weighting scheme for calculating the Kappa coefficient is linear weighting. The acute stress index is defined as the membership degree of acute stress type in the system output multiplied by 100, and the developmental lag index is the membership degree of chronic developmental lag type multiplied by 100. Both range from 0 to 100. The ratio of the two is introduced as a comprehensive disease course discrimination factor, where acute stress... When both the stress index and the developmental lag index are non-zero, the ratio is calculated, and the discrimination rules are as follows: If the ratio is greater than 2.0 and U-shaped metabolites dominate in the metabolic trajectory (the proportion of U-shaped metabolites to the total number of characteristic metabolites exceeds 60%), it is judged as acute stress injury; if the ratio is less than 0.5 and monotonically decreasing characteristic metabolites dominate (the proportion of monotonically decreasing metabolites exceeds 60%), it is judged as chronic developmental lag; if the ratio is between 0.5 and 2.0 or neither type of metabolite reaches a dominant proportion (both are below 60%), it is judged as mixed injury. The discrimination results, along with the acute stress index, developmental lag index, and the proportion of metabolite categories, are written into memory for subsequent use, and the names of the main contributing metabolites that led to the discrimination and their temporal characteristics are also marked.

[0066] Step 3: Overlay deep Q-network with real-time streaming data from wearable multi-parameter biochemical sensors to dynamically extrapolate the intestinal epithelial repair threshold and predict whether the current intestinal microenvironment supports epithelial function reconstruction. This quantitative prediction of intestinal epithelial repair conditions provides a scientific basis for intervention decisions. The wearable multi-parameter biochemical sensor is fixed to the subcutaneous tissue behind the ear of piglets. Continuous streaming data of intestinal fatty acid-binding protein, pH, lactic acid, and glucose concentrations are collected using a variable-interval sampling strategy. Kalman filtering is used to remove motion artifacts, achieving non-invasive continuous monitoring, effectively filtering out motion interference, and obtaining stable and reliable physiological parameters. Time-series data, together with current sensor readings and acquired damage location and disease course information, are used to construct an environmental state vector. The action space of the deep Q network is defined to include three types of repair actions: promoting proliferation, anti-inflammatory regulation, and resting observation. Multi-source information is integrated to construct a state representation, providing a complete environmental perception basis for intelligent decision-making. The in vitro migration rate of crypt cells and the intestinal epithelial organoid formation rate are set as reward signals. The value of actions is explored through an ε-greedy strategy. The Q network weights are iteratively updated until the state-action value function converges, driving the model to autonomously optimize the intervention strategy and achieve learning convergence from data to optimal repair decision.

[0067] It should be noted that the wearable multi-parameter biochemical sensor integrates a microneedle array for detecting intestinal fatty acid-binding protein, a pH glass electrode, a lactate oxidase electrode, and a glucose oxidase electrode. The sensor thickness is no more than 3mm, and its weight is less than 5g. It is fixed by medical-grade silicone. The sampling strategy adopts a variable interval sampling: a 60-minute sampling interval during nighttime (20:00 to 6:00 the next day), and an increased interval of 30 minutes during peak daytime activity (6:00 to 20:00). The raw signal is converted to digital using a 12-bit analog-to-digital converter and then input into a Kalman filter. The initial value of the noise covariance matrix was set to 0.1 times the identity matrix. The measurement noise covariance matrix was set according to the factory calibration accuracy of each sensor. The intestinal fatty acid binding protein was 0.2 ng / mL, pH was 0.05, lactate was 0.1 mmol / L, and glucose was 0.1 mmol / L. The filter output was smoothed time-series data after outlier removal. Resampling calibration was triggered when the signal of two consecutive sampling points deviated from the predicted value by more than 3 times the standard deviation of the measurement noise. The current sensor filter output value was combined with the mean probability of damage from the damage heatmap and the primary damage. The percentage of injured area, along with the acute stress index and developmental lag index output from step 2, are used to construct a ten-dimensional environmental state vector. In this vector, intestinal fatty acid-binding protein values ​​range from 0.5 to 5.0 ng / mL, pH from 6.0 to 7.5, lactate from 0.5 to 5.0 mmol / L, glucose from 2.0 to 6.0 mmol / L, the mean damage probability is 0-1, the percentage of injured area is 0-100%, and both indices are 0-100. The deep Q-network action space is defined as three types of repair actions: promoting proliferation by upregulating pancreatic islets... The cytokinin-like growth factor-1 pathway enhances crypt cell division; anti-inflammatory regulatory actions inhibit the nuclear factor κB signaling pathway and reduce the release of inflammatory factors; resting observation actions involve no intervention but continuous monitoring. These three types of actions are mutually exclusive and discrete, with only one action executed within a single decision cycle, synchronized with the sensor sampling cycle. The reward signal consists of two components: crypt cell migration rate in vitro and intestinal epithelial organoid formation rate. The crypt cell migration rate is indirectly characterized by the negative change rate of intestinal fatty acid-binding protein levels during continuous sampling periods. The expression for the crypt cell migration rate reward is: , The reward for crypt cell migration rate is expressed in ng / mL·h, with an upper limit of 0.5. The difference between adjacent decision times. The organoid formation rate reward is inferred based on the sensor pH and lactate-glucose ratio. When pH ≥ 7.2 and lactate / glucose ratio ≤ 0.5, a positive reward of 0.3 is given; otherwise, the reward is 0. The ε-greedy strategy has an initial exploration rate of 0.9, which decays to 0.95 after every 1000 decisions. The minimum exploration rate is 0.05. The Q network adopts a three-layer fully connected structure with 128, 64, and 32 hidden neurons, respectively. The learning rate is 0.001, the target network update step size is 100 steps, the experience replay pool capacity is 10,000 sets, and the training termination condition is that the average reward change is less than 5% for 500 consecutive rounds, and the state-action value function fluctuates within 0.1 times the initial value on the validation set.

[0068] Furthermore, step 3 also includes: extracting the optimal action value function from the converged deep Q-network model, back-calculating the maximum threshold of intestinal epithelial proliferation corresponding to the repair action under various intestinal microenvironment states, constructing a repair threshold surface, realizing a quantitative mapping between repair conditions and action effects, improving the scientific nature of threshold setting, mapping the current real-time readings of wearable sensors to the corresponding coordinate points of the threshold surface, calculating the Euclidean distance of each continuous parameter from the surface boundary, dynamically quantifying the minimum energy substrate concentration and maximum inflammatory tolerance level required to maintain epithelial reconstruction, realizing real-time quantitative assessment of the microenvironment state, providing dynamic basis for intervention decisions, and outputting a three-segment repair threshold range based on short-chain fatty acids, endotoxin concentration, and redox potential, corresponding to three microenvironment levels: high risk, critically repairable, and supportive of repair, respectively, intuitively classifying the microenvironment repair level, and assisting in quickly judging the urgency of intervention;

[0069] It should be noted that, using a ten-dimensional environmental state vector (intestinal fatty acid-binding protein, pH, lactate, glucose, mean damage probability, damage area percentage, acute stress index, developmental lag index, and the ratio derived from the two indices) as input coordinates, the maximum threshold of intestinal epithelial proliferation rate corresponding to the execution of three types of repair actions under various microenvironmental states is derived. The threshold surface is constructed using a grid search method, discretizing the continuous parameters (lactic acid 0.5-5.0 mmol / L, glucose 2.0-6.0 mmol / L, pH 6.0-7.5) with a step size of 0.1, and using the mean damage probability and area percentage as weighting coefficients, the boundary of the metabolite combination corresponding to the maximum Q value when performing proliferative actions is calculated. This method aims to determine the minimum energy substrate concentration required to maintain crypt cell proliferation, specifically, the product of lactate and glucose concentrations must be no less than 2.5 mmol / L². Simultaneously, it infers that the upper boundary value of intestinal fatty acid-binding protein concentration is 2.8 ng / mL under conditions where anti-inflammatory regulation is effective. Exceeding this value results in inflammation levels exceeding the range of anti-inflammatory regulation, hindering epithelial remodeling. Real-time readings from wearable sensors are smoothed using Kalman filtering and mapped to the corresponding coordinate points on the repair threshold surface using four-dimensional parameters (lactic acid, glucose, pH, and intestinal fatty acid-binding protein). The Euclidean distance of each parameter from the surface boundary is calculated. Specifically, the product of lactate and glucose is calculated relative to 2.5 ng / mL. The deviations in the ratio of lactate to glucose (mmol / L)² threshold and the deviations in the difference between intestinal fatty acid-binding protein and 2.8 ng / mL are used to calculate the root mean square of these two deviations as the comprehensive distance. If the comprehensive distance is less than 0.2, the microenvironment is considered to be at the edge of the critical repairable range. If the current pH is below 6.5 or the product of lactate and glucose is less than 1.8 (mmol / L)², and the distance is greater than 0.5, it indicates severe energy substrate deficiency or acidosis, and the conditions for epithelial remodeling are not met. The dynamic quantitative results are updated to memory in real time, outputting the minimum energy substrate concentration (lactate to glucose product ≥ 2.0 (mmol / L)² is the minimum acceptable value) and the maximum inflammatory tolerance level (intestinal fatty acid-binding protein ≥ 2.8 ng / mL) required to maintain epithelial remodeling. (The upper limit threshold is ≤3.0 ng / mL for acid-binding protein). In the three-stage repair threshold range, the concentration of short-chain fatty acids is inversely estimated by the product of lactate and glucose and pH. When the product is ≥2.5 (mmol / L)² and pH is ≥7.0, the estimated equivalent concentration of butyrate is ≥8 mmol / L. The endotoxin concentration is characterized by the level of intestinal fatty acid-binding protein. ≤2.0 ng / mL is set as the supportive repair range (corresponding to endotoxin ≤0.3 EU / mL), 2.0-2.8 ng / mL is the critical repairable range, and >2.8 ng / mL is the high-risk range. The redox potential is estimated by the lactate / glucose ratio and pH. The lactate / glucose ratio is ≤0.4 and pH is ≥7.2. Estimated potential ≤ -80mV (reduced state), three-stage classification: if all three parameters reach the support repair range, it is determined to be at the support repair level; if any one parameter is in the critical range and there are no high-risk parameters, it is determined to be critically repairable; if any one parameter reaches the high-risk range, it is determined to be at the high-risk level.

[0070] Step 4: Based on the estimated repair threshold, assess whether the current gut microenvironment supports epithelial reconstruction, predict the adaptive recovery potential of piglet gut development and function, accurately assess the microenvironment's support for epithelial reconstruction, and improve the accuracy of recovery potential prediction.

[0071] Step 5: Based on the prediction results, adaptively trigger supplementary feeding, enzyme preparation or microbial intervention instructions to achieve individualized regulation of intestinal development and function detection in weaned piglets, improve the efficiency of intestinal function recovery in weaned piglets, achieve individualized and precise intervention and regulation, and significantly improve the efficiency of intestinal function recovery.

[0072] Step 6: Integrate location, time series, prediction and intervention data to generate a report on the intestinal development and function of weaned piglets. This enables a full-chain assessment from the identification of insidious damage to the prediction of functional adaptation, supporting precision breeding management and forming a credible full-chain assessment report to provide complete decision support for precision breeding management.

[0073] Example 2, as Figure 1 , Figure 2 As shown, based on Example 1, the present invention provides a technical solution: Step 4 specifically includes: collecting real-time values ​​of three key parameters representing energy metabolism (lactic acid and glucose), barrier integrity (intestinal fatty acid binding protein), and inflammatory state (real-time indirectly calculated pH fluctuations) from wearable sensors at the current moment, realizing multimodal synchronous capture of key microenvironment indicators, comparing the real-time values ​​of the three key parameters with the corresponding dimension repair thresholds, and determining high recovery potential if all three parameters simultaneously meet the conditions of being better than or reaching the support repair range, and determining low recovery potential if any one of them is lower than the critical repair lower limit, thereby realizing the immediate classification and determination of microenvironment support capability. For the low recovery potential determination, the parameter dimension that first breaks through the lower limit is further identified as the bottleneck factor, and the bottleneck factor type label and the quantitative difference of deviation from the threshold are output, thereby realizing the precise positioning of the key obstacles restricting repair.

[0074] It should be noted that real-time data from a wearable multi-parameter biochemical sensor, smoothed by Kalman filtering, was collected at the current moment. Three key parameters representing energy metabolism, barrier integrity, and inflammatory status were extracted. Energy metabolism parameters included lactate and glucose concentrations, with ranges of 0.5-5.0 mmol / L and 2.0-6.0 mmol / L, respectively. Sampling intervals were 30 minutes during the day and 60 minutes at night. Real-time values ​​were output by a filter after 12-bit analog-to-digital conversion. The barrier integrity parameter was the concentration of intestinal fatty acid-binding protein, with a detection range of 0.5-5.0 ng / mL, measured using a microneedle array combined with enzyme-linked immunosorbent assay (ELISA), with a response time ≤5 minutes. The inflammatory status parameter used was real-time indirectly calculated pH fluctuations, detected by a glass electrode. The pH range is 6.0-7.5, with a resolution of 0.01. The level of inflammatory activity is characterized by the pH change between two consecutive sampling points. Inflammation is considered activated when the pH decrease rate exceeds 0.05 / hour and occurs twice consecutively. Real-time values ​​of the three parameters are updated to memory after each sensor sampling. The real-time values ​​of the three key parameters are compared with corresponding repair thresholds: the energy metabolism repair threshold is defined as the product of lactate and glucose concentrations not less than 2.5 (mmol / L)²; the barrier integrity repair threshold is defined as the concentration of intestinal fatty acid-binding protein not exceeding 2.8 ng / mL; and the inflammatory state repair threshold is defined as pH not lower than 6.5 and pH fluctuation not exceeding 0.1 / hour. The judgment rule is as follows: if all three parameters are simultaneously satisfied, the condition is considered superior. If the gut microenvironment reaches the aforementioned supportive repair range (lactate-glucose product ≥ 2.5 mmol / L², intestinal fatty acid-binding protein ≤ 2.8 ng / mL, pH ≥ 6.5 with fluctuation ≤ 0.1 mmol / h), it is considered to have high recovery potential, indicating that the current gut microenvironment can support epithelial function reconstruction without external intervention. If any parameter is below the critical repair limit (lactate-glucose product < 2.0 mmol / L², intestinal fatty acid-binding protein > 3.0 ng / mL, or pH < 6.5), it is considered to have low recovery potential, indicating that the current microenvironment has key obstacles restricting repair, requiring further identification of bottleneck factors and intervention. For cases classified as low recovery potential, further identification of the first to break through the critical repair limit is needed. The bottleneck factors are identified as follows: The real-time values ​​of the three types of parameters are compared with their respective critical lower limits in chronological order of sampling time. Parameters that appear for the first time and persist for more than one sampling interval are marked as bottleneck factors. Energy metabolism bottleneck factors are represented by the quantitative difference between the product of lactate and glucose and the threshold, calculated as: Deviation Difference = 2.0 - Current Product Value. A larger deviation indicates a more severe deficiency of energy substrates. Barrier integrity bottleneck factors are represented by the quantitative difference between the concentration of intestinal fatty acid-binding proteins exceeding the upper limit, calculated as: Deviation Difference = Current Concentration - 3.0, reflecting the severity of intestinal epithelial barrier defects. Inflammatory state bottleneck factors are represented by the quantitative difference between pH and the lower limit, calculated as: Deviation Difference = 6.5 - Current pH value, output bottleneck factor type label (energy metabolism bottleneck, barrier integrity bottleneck, or inflammatory state bottleneck) and its quantitative difference from the threshold, for use in matching intervention strategies;

[0075] Step 5 specifically includes: constructing an intervention strategy library, in which energy bottlenecks are matched with supplementary feeding instructions (quantitative addition of lactose or medium-chain triglycerides), digestive bottlenecks are matched with enzyme preparation instructions (combination ratio of protease and amylase), and microbial bottlenecks are matched with microbial intervention instructions (combination of lactic acid bacteria and Bacillus). For each bottleneck factor, two or more compensatory paths and switching priorities are preset. When the preferred path fails due to coexistence constraints, it automatically degrades to the secondary path, ensuring that different bottlenecks have targeted intervention plans. Automatic path switching ensures execution reliability. The system receives the output bottleneck factor type label and the quantitative difference of deviation from the threshold, calls the matching instructions, and calculates the initial intervention dose based on the deviation from the threshold difference. A three-day step-down dosing regimen is determined using a lookup table method to achieve precise matching of dose and damage level. Step-down avoids over-intervention and promotes smooth recovery. The final generated intervention instructions are encoded into standard aquaculture management messages, pushed to the automated feeding system via wireless network, and locked to prevent repeated triggering within the execution cycle. Simultaneously, the instruction trigger time and parameters are recorded for subsequent reporting, achieving automated closed-loop transmission and anti-repeated execution of intervention instructions, supporting full-process traceability and effect evaluation.

[0076] It should be noted that an intervention strategy library containing three types of bottleneck factors was pre-constructed. For energy metabolism bottlenecks, supplementary feeding instructions were matched, with the preferred compensatory pathway being lactose supplementation at a dose of 0.5g per kilogram of body weight, dissolved in drinking water and administered twice daily. The secondary pathway was medium-chain triglyceride supplementation at a dose of 0.3mL per kilogram of body weight, mixed with feed and administered once. For barrier integrity bottlenecks, barrier repair supplementary feeding instructions were matched, with the preferred pathway being glutamine supplementation at a dose of 0.2g per kilogram of body weight. The secondary pathway was zinc supplementation (zinc methionine) at a dose of 15mg daily. For digestive enzyme activity bottlenecks, enzyme preparation instructions were matched, with the preferred pathway being a 1:2 ratio of protease and amylase, with an addition of 500U protease and 1000U amylase per kilogram of feed. The secondary pathway was adding protease alone (800U per kilogram of feed). For gut microbiota dysbiosis bottlenecks, microbial intervention instructions were matched, with the preferred pathway being a 1:1 ratio of *Lactobacillus plantarum* and *Bacillus subtilis*, with a total daily live bacteria count of 1×10⁻⁶. 9 CFU, secondary route of administration is single-strain *Saccharomyces boulardii*, 5 x 10 times daily. 9For each bottleneck factor, CFU (Continuous Functional Fusion) sets two compensatory pathways and establishes a switching priority queue. When the preferred pathway is deemed unfeasible due to coexisting constraints such as piglet refusal to eat, coexisting drug treatment, or existing similar interventions, it automatically downgrades to the secondary pathway and records the reason for the pathway switch. After receiving the bottleneck factor type label and the deviation threshold quantification difference, it calls the matching instruction and calculates the initial intervention dose. The deviation difference of the energy metabolism bottleneck is expressed as a value where the product of lactate and glucose is less than 2.0 (mmol / L)². For every 0.1 increase in deviation, the base dose increases by 10%. The upper limit of the initial lactose dose is 1.0g per kilogram of body weight. The deviation difference of the barrier integrity bottleneck is I-FABP exceeding... The target value is 3.0 ng / mL. For every 0.2 ng / mL deviation, the glutamine dose is increased by 20%, with an upper limit of 0.4 g / kg body weight. The bottleneck for inflammatory states is defined as a pH value below 6.5. For every 0.1 ng / mL deviation, the dose is corrected by supplementing with an alkalizing agent (sodium citrate), starting at 0.1 g / kg body weight. A step-down dosing regimen for three consecutive days is determined using a lookup table: Day 1 is 100% of the calculated initial dose, Day 2 is adjusted to 70%, and Day 3 is adjusted to 50%. For mild bottlenecks with a deviation of less than 0.2 ng / mL, the reduction regimen is shortened to two days (100%, 60%). For severe bottlenecks with a deviation of greater than 0.8 ng / mL, the dose is reduced to 100% and 60% for the first three days. After completion, a 30% maintenance dose is added on the fourth day as a consolidation treatment. The standard breeding management message format includes a message header, piglet electronic ear tags, instruction type codes (supplementary feeding instruction 0x01, enzyme preparation instruction 0x02, microbial intervention instruction 0x03), bottleneck factor type codes (0x10 energy, 0x11 barrier, 0x12 digestive enzymes, 0x13 microbiota), a three-day decreasing dose sequence (units are g, U or CFU, each occupying 4 bytes), execution time period (start timestamp and daily execution window, each occupying 8 bytes), and a CRC32 checksum tail. The total message length does not exceed 128 bytes and is pushed to the automated feeding system via a 2.4GHz wireless network. The central controller, after parsing, triggers the corresponding feed bin or liquid replenishment device to execute. During the instruction execution cycle, it locks the intervention task corresponding to the piglet's electronic ear tag. Even if the same bottleneck factor trigger signal is received again during the locking period, the intervention instruction will not be generated or pushed again. At the same time, the instruction trigger time, bottleneck factor type, deviation difference, selected compensation path and execution status are written to the local SQLite database. The fields include trigger timestamp, ear tag number, instruction type, dosage sequence, execution completion flag and abnormal record. If the automated feeding system pushes three times in a row without response, it will be downgraded to push to the feeder's handheld terminal, where manual intervention will be performed and the execution confirmation information will be scanned and sent back.

[0077] Step 6 specifically includes: integrating the output damage heatmap time-series snapshot, metabolic trajectory and disease course classification labels, repair threshold intervals and potential grading and intervention records, constructing a multi-source heterogeneous dataset, realizing unified management and rapid retrieval of multi-dimensional detection data, and performing report template filling. The damage heatmap is presented as a color spatial distribution map, the metabolic trajectory is presented as an age-abundance line graph with fuzzy system discrimination markers, and the repair threshold interval is displayed in three dimensions in the form of a three-dimensional surface, intuitively presenting the spatial relationship between damage evolution, metabolic trajectory and repair conditions. A conclusion paragraph is generated in the report, arranged in the order of localization results, time-series attribution, current repair conditions, intervention suggestions and reassessment time points. The reassessment time points are dynamically predicted by the deep Q network based on the current repair speed rather than a fixed period. All original data are attached with anti-tampering hash verification codes and QR code index links, providing dynamic assessment cycles and ensuring the reliable traceability of the entire data chain.

[0078] It should be noted that after completing all detection and intervention, the following data was integrated: time-series snapshots of the injury heatmap (including three-dimensional intestinal anatomical coordinate system injury probability distribution maps on days 0, 3, 7, 10, and 14, Jet color mapping, with primary areas marked in red and secondary areas marked in orange), metabolic trajectory and disease course classification labels (including corrected abundance curves of each characteristic metabolite at 6 time points, superimposed with 95% confidence bands of age-matched healthy piglets, with inflection points marked by triangles and fluctuation intervals marked by diamond boxes), repair threshold intervals (a three-dimensional surface with short-chain fatty acid equivalent concentration, estimated endotoxin concentration, and redox potential as coordinate axes, with a grid step size of 0.1), and potential grading and intervention records (including high / critical / low recovery). The report template uses a three-column layout. The damage heatmap is arranged horizontally along the time axis at the beginning of the report. Each snapshot is labeled with the percentage of damage area and the coordinates of the centroid of the largest damage cluster. The metabolic trajectory uses age as the x-axis and normalized abundance as the y-axis. Each metabolite curve is distinguished by a different color. Areas exceeding the 95% confidence band are highlighted with a gray background, inflection points are marked with black triangles, and fluctuation ranges are covered with yellow semi-transparent bands. The repair threshold surface is displayed as a three-dimensional surface map, showing the current microenvironment of the piglets. Status points are marked with red spheres in the curved space, and the contour lines of the curved surface are projected onto the bottom surface to aid interpretation. The report conclusions section is arranged in five subsections: localization results, temporal attribution, current repair conditions, intervention recommendations, and reassessment time points. The localization results section summarizes the proportion of damage distribution in each intestinal segment, the spatial coordinate range of primary and secondary regions, and the maximum diameter length. The temporal attribution section outputs the course type (acute stress injury, chronic developmental lag, or mixed type) determined by attention-based ANFIS, and lists the specific values ​​and ratios of the acute stress index and the developmental lag index, while also indicating the names and temporal characteristics of the main contributing metabolites that led to the determination. The current repair conditions section is based on a three-stage repair level (branch... The system assesses the recovery potential level (repairable / critically repairable / high-risk), describes the deviation of the current energy metabolism product, intestinal fatty acid-binding protein concentration, and pH value from the repair threshold, and lists specific plans for supplemental feeding, enzyme preparations, or microbial interventions, along with stepwise dose reductions, based on the actual triggered instruction type. The reassessment time point is dynamically predicted by the deep Q network based on the current repair rate. The calculation formula is that the next assessment interval equals 24 hours divided by one plus the repair rate coefficient, where the repair rate coefficient is calculated by weighting the slope of improvement in the energy metabolism product and the slope of decrease in intestinal fatty acid-binding protein over 6 consecutive hours. The dynamic prediction results are output in hourly units and are not at a fixed period.The report includes tamper-proof hash verification codes and QR code index links for all raw data to ensure the traceability and non-repudiation of the test results. The hash verification uses the SHA-256 algorithm, performing concatenated hash calculations on the methylation quantification and fluorescence intensity time series matrix, metabolomics time series fingerprint database, wearable sensor filter output sequences, potential grading records, and intervention instruction logs to generate a fixed-length 64-byte hexadecimal string, printed in the report footer. The QR code index link points to a password-protected cloud data storage server; scanning allows for sequential viewing along the timeline, including continuous sampling curves of intestinal fatty acid-binding protein and lactate glucose concentrations (including Kalman filters). The report includes: pre- and post-wave comparisons; raw data tables of corrected abundance for each metabolite and marked records exceeding the 95% confidence band; an interactive browsing model of a 3D damage heatmap (supporting switching between coronal, sagittal, and transverse views); timestamp logs of intervention command pushes and execution confirmations; and server-side storage of system version numbers and algorithm parameter snapshots for subsequent auditing or retrospective analysis. The report notes explain the hash verification value calculation method and QR code usage permissions. Aquaculture managers can scan the QR code using a dedicated terminal to verify data integrity. Any tampering with the raw data will result in hash verification failure, ensuring the reliability of the comprehensive detection report across the entire chain from damage localization to intervention execution.

[0079] The following is a detailed description of the workflow of a comprehensive detection method for intestinal development and function in weaned piglets.

[0080] First, fresh fecal samples were collected from weaned piglets. DNA methylation markers were extracted from the promoter region of tight junction protein genes in exfoliated intestinal epithelial cells. Simultaneously, near-infrared fluorescent nanosensors were orally administered. After the nanosensors distributed in the intestine, spectral signals were collected through the abdominal wall. The methylation data and spectral features were spatiotemporally registered, and a pixel-level intestinal injury probability map was generated using a convolutional neural network. A three-dimensional injury heatmap was output, and primary and secondary injury areas were distinguished. Based on this, contents of the terminal ileum were collected at multiple post-weaning time points. The temporal abundance data of short-chain fatty acids, tryptophan metabolites, and secondary bile acids were obtained by ultra-high performance liquid chromatography-tandem mass spectrometry. After tissue background correction, a metabolomics temporal map was constructed. The temporal data was input into an adaptive neurofuzzy inference system that incorporates an attention mechanism. Semi-supervised training was performed using pseudo-labels generated by temporal hierarchical clustering. The acute stress index and developmental lag index were output to determine whether the injury was acute stress type, chronic developmental lag type, or mixed type.

[0081] Subsequently, a wearable multi-parameter biochemical sensor was fixed behind the ear of the piglet. Intestinal fatty acid-binding protein, pH, lactate, and glucose concentrations were continuously collected using a variable-interval strategy. After Kalman filtering and smoothing, these data, along with the aforementioned damage localization and disease progression information, were used to construct an environmental state vector. This vector was then input into a deep Q-network, which was trained using crypt cell migration rate and organoid formation rate as reward signals. The resulting repair threshold surface was then derived, dynamically quantifying the energy substrate concentration and inflammatory tolerance level required to maintain epithelial remodeling. A three-stage repair level was output, and the real-time sensor parameters were compared with the repair threshold to determine high or low recovery potential. Low recovery potential was then identified. The bottleneck factor type and deviation difference are used to calculate the stepwise dose reduction instructions in the supplementary feeding, enzyme preparation or microbial intervention instructions in the bottleneck factor matching intervention strategy library and encode them into a standard message, which is pushed to the automated feeding system for execution and lock the cycle. Finally, all data are integrated to generate a comprehensive test report, including a time-series snapshot of the damage heat map, metabolic trajectory and disease course labels, repair threshold surface and intervention records. The conclusion paragraph is arranged in the order of localization results, time-series attribution, repair conditions, intervention suggestions and dynamic prediction re-evaluation time points, and is attached with hash check code and QR code index link to realize the whole chain evaluation from the identification of insidious damage to the prediction of functional adaptation.

[0082] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A comprehensive detection method for intestinal development and function in weaned piglets, characterized in that, Includes the following steps: Step 1: Collect near-infrared spectra of feces and abdominal wall of weaned piglets, and fuse DNA methylation markers and near-infrared fluorescent nanosensors to non-invasively locate intestinal micro-damage areas; Step 2: Collect intestinal contents of weaned piglets at different ages, construct metabolomic time series maps, integrate an adaptive neurofuzzy reasoning system, retrospectively analyze the acute or chronic course of the located damage, and distinguish between stress-induced damage and developmental delay. Step 3: Overlay deep Q-network with real-time streaming data from wearable multi-parameter biochemical sensors to dynamically extrapolate the intestinal epithelial repair threshold and predict whether the current intestinal microenvironment supports epithelial function reconstruction; Step 4: Based on the estimated repair threshold, assess whether the current gut microenvironment supports epithelial reconstruction and predict the adaptive recovery potential of piglet gut development and function. Step 5: Based on the prediction results, adaptively trigger instructions for supplemental feeding, enzyme preparation, or microbial intervention; Step 6: Integrate location, time series, prediction and intervention data to generate a report on the intestinal development and function of weaned piglets.

2. The comprehensive detection method for intestinal development and function of weaned piglets according to claim 1, characterized in that: Step 1 specifically includes: Fresh fecal samples were collected from weaned piglets. The methylation level of CpG islands in the promoter regions of the tight junction proteins Occludin and Claudin-1 in the fecal exfoliated intestinal epithelial cells was extracted using a magnetic bead capture method. The spatial distribution signals of intestinal permeability and inflammatory activity were simultaneously obtained by combining the samples with a near-infrared fluorescent nanosensor excited by transmission through the abdominal wall after mixing with the feces. The acquired quantitative methylation data was spatiotemporally registered with the near-infrared spectral characteristic peak intensities corresponding to different intestinal segments. A methylation-spectral joint feature map was constructed using a convolutional neural network, thereby generating a pixel-level intestinal injury probability map. Based on the generated intestinal injury probability map, an adaptive confidence threshold that balances positive predictive value and false positive rate is set. The spatial coordinates and ranges of high-probability cryptic micro-injury areas are automatically segmented and labeled from the map, and a three-dimensional injury heat map with intestinal anatomical location as reference is output.

3. The comprehensive detection method for intestinal development and function of weaned piglets according to claim 2, characterized in that: Step 1 further includes: Near-infrared spectral time-series data of the abdominal wall of the same piglet were collected on days 0, 3, 7, 10 and 14 after weaning. Fluorescence intensity change curves of projection points on each body surface were constructed to dynamically track the spatial migration trajectory of the near-infrared fluorescent nanosensor signal. The methylation detection results were stratified by time point, and the consistency between the spatial distribution of DNA methylation spectrum and fluorescence signal at different ages was compared using a dynamic time warping algorithm to identify whether the damaged area has expanded or migrated. Based on the direction of signal migration and the rate of change in methylation, the initial site of insidious damage and the secondary diffusion path can be located, and the spatial range of primary and secondary damage can be distinguished.

4. The comprehensive detection method for intestinal development and function of weaned piglets according to claim 1, characterized in that: Step 2 specifically includes: The contents of the terminal ileum of piglets were collected on days 0, 2, 5, 9, 14 and 21 after weaning. The relative abundance of small molecule metabolites was obtained by ultra-high performance liquid chromatography-tandem mass spectrometry. The count of exfoliated cells in the intestinal mucosa in each sample was recorded as a tissue background correction parameter for the source of metabolites. We used sparse partial least squares discriminant analysis based on elastic network regularization to reduce the dimensionality of time-series metabolic data, and automatically screened out three types of characteristic metabolite combinations that showed monotonically increasing, monotonically decreasing, or U-shaped changes with the weaning age, and constructed a metabolomics time-series map with age as the horizontal axis. The rate of change of each characteristic metabolite at adjacent time points was calculated. Combined with the dynamic change curve of intestinal mucosal exfoliated cell count, the inflection points and fluctuation ranges in the metabolic trajectory that exceed the 95% confidence interval of healthy piglets of the same age were marked, forming a metabolomic time-series fingerprint database that reflects the dynamic evolution of intestinal mucosal barrier function.

5. A comprehensive detection method for intestinal development and function in weaned piglets according to claim 4, characterized in that: Step 2 also includes: The time-series curves of each metabolite in the metabolome time-series fingerprint database are input into an adaptive neurofuzzy inference system that integrates an attention mechanism. The time point is used as the antecedent input and the damage course type is used as the consequent output. Prior knowledge of the causal relationship between metabolites is introduced as the initial weight of the fuzzy rules, and three fuzzy membership functions are set. The pseudo-labels generated by time-series hierarchical clustering were used for semi-supervised pre-training to initialize the fuzzy rule base. The rule parameters were adjusted by a hybrid learning algorithm that alternately optimized backpropagation and least squares estimation, so as to maximize the weighted Kappa coefficient of the system output discrimination result and the actual intestinal mucosal histology score of the piglet on the 28th day. The system outputs an acute stress index and a developmental lag index after training. The ratio of the two is introduced as a comprehensive disease course discrimination factor. If the factor is greater than 2.0 and U-shaped characteristic metabolites dominate in the metabolic trajectory, it is judged as acute stress injury. If the factor is less than 0.5 and monotonically decreasing characteristic metabolites dominate, it is judged as chronic developmental lag. The boundary interval is judged as mixed injury.

6. The comprehensive detection method for intestinal development and function of weaned piglets according to claim 1, characterized in that: Step 3 specifically includes: Wearable multi-parameter biochemical sensors were fixed to the subcutaneous tissue behind the ears of piglets. Continuous streaming data of intestinal fatty acid binding protein, pH value, lactic acid and glucose concentration were collected using a variable interval sampling strategy, and Kalman filtering was used to remove motion artifacts. An environmental state vector is constructed by combining current sensor readings with the obtained damage location and disease course information. The action space of the deep Q network is defined to include three types of repair actions: promoting proliferation, anti-inflammatory regulation, and resting observation. The in vitro migration rate of crypt cells and the intestinal epithelial organoid formation rate are set as reward signals. The value of actions is explored through an ε-greedy strategy, and the weights of the Q network are iteratively updated until the state-action value function converges.

7. A comprehensive detection method for intestinal development and function in weaned piglets according to claim 6, characterized in that: Step 3 also includes: The optimal action value function is extracted from the converged deep Q-network model, and the maximum threshold of intestinal epithelial proliferation rate corresponding to the execution of repair actions under various intestinal microenvironment states is deduced to construct the repair threshold surface. The current real-time readings of the wearable sensor are mapped to the corresponding coordinate points of the threshold surface, and the Euclidean distance of each continuity parameter from the surface boundary is calculated to dynamically quantify the minimum energy substrate concentration and maximum inflammatory tolerance level required to maintain epithelial regeneration. The output is a three-stage repair threshold range based on short-chain fatty acid, endotoxin concentration, and redox potential, corresponding to three microenvironment levels: high risk, critically repairable, and supportive of repair.

8. A comprehensive detection method for intestinal development and function in weaned piglets according to claim 1, characterized in that: Step 4 specifically includes: Collect real-time values ​​of three key parameters representing energy metabolism, barrier integrity, and inflammatory status from wearable sensors at the current moment; The real-time values ​​of the three key parameters are compared with the corresponding dimension repair thresholds. If all three parameters are simultaneously better than or reach the support repair range, they are judged to have high recovery potential. If any one of them is lower than the critical repairable lower limit, it is judged to have low recovery potential. For the determination of low recovery potential, the parameter dimension that first breaks through the lower limit is further identified as the bottleneck factor, and the bottleneck factor type label and the quantitative difference of deviation from the threshold are output.

9. A comprehensive detection method for intestinal development and function in weaned piglets according to claim 1, characterized in that: Step 5 specifically includes: An intervention strategy library is constructed, in which energy bottlenecks are matched with supplementary feeding instructions, digestion bottlenecks are matched with enzyme preparation instructions, and microbial bottlenecks are matched with microbial intervention instructions. For each bottleneck factor, two or more compensatory paths and switching priorities are preset. When the preferred path fails due to coexistence constraints, it is automatically downgraded to the secondary path. Receive the bottleneck factor type label and the quantitative difference of the deviation threshold from the output, call the matching instruction and calculate the initial intervention dose based on the difference of the deviation threshold, and determine the step-down dosing regimen for three consecutive days by looking up the table. The final generated intervention command is encoded into a standard aquaculture management message, pushed to the automated feeding system via wireless network, and locked to be triggered repeatedly within the execution cycle. At the same time, the command trigger time and parameters are recorded.

10. A comprehensive detection method for intestinal development and function in weaned piglets according to claim 1, characterized in that: Step 6 specifically includes: By integrating the output damage heatmap time-series snapshots, metabolic trajectories and disease course classification labels, repair threshold ranges and potential grading and intervention records, a multi-source heterogeneous dataset is constructed. The execution report is templated and populated, with damage heatmaps presented as color spatial distribution maps, metabolic trajectories as age-abundance line graphs overlaid with fuzzy system discrimination markers, and repair threshold intervals displayed in three dimensions as three-dimensional curved surfaces. The report generates a conclusion paragraph, arranged in the order of location results, time-series attribution, current remediation conditions, intervention recommendations, and reassessment time points, and includes tamper-proof hash verification codes and QR code index links for all original data.

Images