A forage grass drying evaluation method and system based on multi-modal fusion

Abstract

The present application relates to the technical field of forage drying evaluation, and discloses a forage drying evaluation method and system based on multi-modal fusion. The method establishes a quality file containing indexes such as moisture content, dry matter content, and dryness level for each batch of forage, and synchronously collects the intake amount, feeding behavior, and production performance data of the cattle herd during feeding. A multi-modal correlation data set is constructed with batch identification and time stamp as the correlation key. Then, correlation analysis, regression analysis, and statistical significance test are used to obtain the feeding performance and production performance response results corresponding to different drying characteristics, and the forage drying evaluation standard is dynamically revised accordingly. The present application correlates the forage drying characteristic data with the intake amount, feeding behavior, and production performance data of the cattle herd by batch and time, constructs a multi-modal correlation data set, and realizes the direct correspondence between the forage drying state and the actual feeding performance and growth effect of the cattle herd.

Description

Technical Field

[0001] This invention belongs to the field of forage drying assessment technology, specifically relating to a forage drying assessment method and system based on multimodal fusion. Background Technology

[0002] In the breeding of Red Cattle in Pingliang, Gansu Province, scientific feeding and management techniques are implemented. Currently, feeding whole-plant corn silage to cattle farms is highly recommended. The degree of dryness of the forage is an important factor affecting the safety and utilization of forage storage. In existing technologies, forage dryness assessment usually uses moisture content or dry matter content as the main evaluation indicators, and sets uniform dryness qualification standards accordingly.

[0003] A search revealed a method and system for evaluating the quality of unconventional feed, disclosed in publication number CN121336932A. The method includes the following steps: during the high-temperature drying process of feed, a multi-dimensional sensing monitoring baseline is established, and data on mercury vapor concentration gradient, moisture evaporation rate, and airflow velocity field in the feed drying environment are continuously acquired to form a time-series feature set reflecting the dynamic changes of mercury vaporization; using the time-series feature set, an analytical model of gas-solid interaction is constructed, and the mercury vapor migration trajectory is coupled with the temperature and humidity distribution on the surface of feed particles to generate a spatial distribution dataset characterizing the spatial distribution features of the mercury vapor redeposition area.

[0004] However, the inventors discovered during actual aquaculture production that the above-mentioned forage drying assessment method based on a single physical indicator still has the following shortcomings:

[0005] The results of forage drying assessment are not correlated with the actual feeding performance of cattle. Existing technology only focuses on the physical drying state of forage and does not take into account the differences in feeding behavior of cattle forage with different degrees of dryness. This can easily lead to situations where the drying indicators are qualified but the cattle have low feed intake and are picky eaters.

[0006] There is a data gap between drying assessment and production performance. Existing technologies do not correlate the results of forage drying assessment with production performance indicators such as average daily weight gain and health status of cattle, resulting in forage quality assessment failing to reflect the true breeding effect.

[0007] Fixed drying standards can easily lead to over-drying and energy waste. In the absence of a feedback mechanism, in order to meet traditional drying standards, it is often necessary to over-dry or over-bake the forage, which not only increases energy costs but may also reduce the palatability of the forage.

[0008] Therefore, there is an urgent need for a forage drying assessment method that can comprehensively analyze forage drying characteristics with cattle feeding behavior and production performance in order to overcome the data silo problem in existing technologies. Summary of the Invention

[0009] The purpose of this invention is to provide a forage drying assessment method and system based on multimodal fusion, in order to solve the problem mentioned in the background art that the existing forage drying assessment is based only on physical drying indicators, which cannot reflect the actual feeding behavior and production performance of cattle, resulting in a disconnect between the drying assessment results and the actual palatability and breeding effect.

[0010] To achieve the above objectives, the present invention provides a forage drying assessment method based on multimodal fusion, comprising the following steps:

[0011] S1. Assign a unique batch identifier to each batch of forage and establish a forage quality file. Collect and record the drying characteristic data of the forage. The drying characteristic data includes, but is not limited to, moisture content, dry matter content, dryness grade and drying uniformity index, and associate it with the time information and batch identifier of the forage.

[0012] S2. During the feeding of the forage batches, the target cattle herd is fed with a variety of feeding data, including at least two of the following: feed intake data, feeding behavior data, and production performance data.

[0013] S3. Based on the batch identifier and timestamp, associate the forage quality file with the breeding data to construct a multimodal association dataset between forage drying characteristics and cattle breeding performance;

[0014] S4. Analyze the multimodal association dataset to obtain the feeding performance and production performance response results corresponding to different forage drying characteristics;

[0015] S5. Based on the feeding performance and production performance response results, make a comprehensive judgment on the forage drying assessment results, and dynamically revise the forage drying assessment standard accordingly, so that the revised forage drying assessment standard can reflect the actual feeding effect and production performance of the cattle herd.

[0016] In one embodiment, in step S1, the forage quality profile further includes forage species information, source plot information, and sampling point information, which are used to characterize the consistency of drying status within the same forage batch.

[0017] In one embodiment, in step S2, the feed intake data includes the total feed intake of the forage batch within a preset time period and the feed intake rate per unit time. The feed intake rate is acquired in real time by a feeding device or feed trough weighing device equipped with a weighing sensor and mapped with the batch identifier.

[0018] In one embodiment, in step S2, the feeding behavior data includes at least one of the following: the herd's feeding activity level, rumination time, chewing frequency, and picky eating behavior characteristics, wherein at least part of the feeding behavior data is acquired by a collar-type behavior sensor and / or a camera image recognition device worn on the cattle.

[0019] In one embodiment, in step S2, the production performance data includes the average daily weight gain (ADG) of the herd during the corresponding forage feeding phase, and may further include health status indicators to characterize adverse reactions.

[0020] In one embodiment, step S4 involves analyzing the multimodal association dataset, including correlation and / or regression analysis between forage drying characteristics and feed intake, feeding behavior, and production performance, to determine the differences in breeding performance corresponding to different drying characteristic intervals.

[0021] In one embodiment, step S5, the dynamic correction of the forage drying assessment criteria includes:

[0022] Based on the forage drying characteristics of increased feed intake, improved production performance, and no abnormal health reactions in cattle, the corresponding drying characteristics are marked as high-quality drying standards.

[0023] Based on the forage drying characteristics of reduced feed intake or decreased production performance in cattle, the corresponding drying characteristics are marked as inferior drying standards.

[0024] In one embodiment, the dynamically modified forage drying assessment criteria are differentiated based on herd type, forage species, or breeding stage to guide the optimization of group feeding, forage prioritization, or forage processing objectives.

[0025] The present invention also provides a forage drying assessment system based on multimodal fusion, comprising:

[0026] The forage data acquisition module is used to collect the drying characteristic data of the forage batch and generate forage quality files;

[0027] The livestock data acquisition module is used to collect feed intake data, feeding behavior data, and production performance data of the corresponding cattle herd during the feeding of the forage batches.

[0028] The data association module is used to associate the forage quality file with the breeding data by batch and time based on the timestamp and the forage batch identifier, and to construct a multimodal association dataset.

[0029] The data storage module is used to store raw data, processed data, and evaluation results indexed by batch identifiers and supports version control.

[0030] The analysis and evaluation module is used to analyze the multimodal association dataset to obtain the feeding performance and production performance response results corresponding to the forage drying characteristics;

[0031] The standard correction module is used to dynamically update the forage drying assessment standard based on the output results of the analysis and evaluation module, and send the updated results to the ranch management system or display interface through the decision output interface.

[0032] The analysis and evaluation module includes a multimodal fusion analysis unit, which is used to comprehensively utilize forage drying characteristic data, feed intake data, feed behavior data and production performance data, and adopt correlation analysis, regression analysis and statistical significance test procedures to comprehensively evaluate the forage drying status and output palatability evaluation results and confidence or uncertainty indicators.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] This invention constructs a multimodal association dataset by associating forage drying characteristic data with cattle feed intake, feeding behavior, and production performance data by batch and time. This achieves a direct correspondence between forage drying status and actual cattle feeding performance and growth effect, thereby overcoming the problem that existing technologies rely solely on physical indicators for forage drying assessment and cannot reflect actual feeding effects.

[0035] By performing correlation analysis, regression analysis, and statistical significance testing on multimodal associated datasets, this invention can obtain the feeding performance and production performance response results corresponding to different forage drying characteristics. This makes the drying assessment results no longer static indicator judgments, but a comprehensive assessment based on actual breeding performance, thereby significantly improving the accuracy and guidance of the assessment results.

[0036] This invention dynamically modifies the forage drying assessment criteria, allowing the assessment targets to be adjusted based on the actual feeding performance and production performance of the cattle herd. When the system determines that the cattle herd still has good feeding performance and production performance under high moisture content conditions, unnecessary further drying treatment can be avoided, thereby reducing energy consumption and lowering the cost of forage processing and storage.

[0037] By differentiating the drying assessment standards according to cattle herd type, breeding stage, or forage species, we can guide different cattle herds to adopt more suitable forage drying ranges, realize group feeding and refined management, and help improve overall breeding efficiency and economic benefits. By evaluating the palatability of forage drying status based on cattle feeding behavior and feed intake, we can help select forage drying characteristic ranges that better match the cattle herd's preferences, reduce feed waste rate, improve forage utilization rate, and further promote the improvement of production performance indicators such as average daily weight gain of cattle. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the process for evaluating forage drying based on multimodal fusion according to the present invention.

[0039] Figure 2 This is a block diagram of the overall structure of a forage drying assessment system based on multimodal fusion according to the present invention.

[0040] Figure 3 This is a schematic diagram illustrating the composition of the forage quality record in Embodiment 1 of the present invention;

[0041] Figure 4 This is a schematic diagram of the multimodal data acquisition method for aquaculture in Embodiment 1 of the present invention;

[0042] Figure 5 This is a schematic diagram of the multimodal correlation data analysis process in Embodiment 1 of the present invention;

[0043] Figure 6 This is a schematic diagram illustrating the dynamic correction and output of the forage drying evaluation standard in Embodiment 1 of the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Example 1: Please refer to Figures 1-6 A method for assessing forage drying based on multimodal fusion includes the following steps:

[0046] S1. Assign a unique batch identifier to each batch of forage and establish a forage quality file. Collect and record the drying characteristic data of the forage, including but not limited to moisture content, dry matter content, dryness grade and drying uniformity index, and associate them with the time information and batch identifier of the forage.

[0047] S2. During the batch feeding of forage, the target cattle herd shall be fed with a variety of feeding data, including at least two of the following: feed intake data, feeding behavior data, and production performance data.

[0048] S3. Based on batch identifiers and timestamps, link forage quality records with breeding data to construct a multimodal association dataset between forage drying characteristics and cattle breeding performance;

[0049] S4. Analyze the multimodal association dataset to obtain the feeding performance and production performance response results corresponding to different forage drying characteristics;

[0050] S5. Based on the feeding performance and production performance response results, comprehensively judge the forage drying assessment results, and dynamically revise the forage drying assessment standards accordingly, so that the revised forage drying assessment standards can reflect the actual feeding effect and production performance of the cattle herd.

[0051] In this embodiment, each batch of forage entering the farm is assigned a unique batch identifier, and a corresponding forage quality file is established on-site.

[0052] The forage quality record should include at least the following: average moisture content (%), dry matter content (%), dryness grade, and drying uniformity index for that batch.

[0053] The on-site measurement of moisture content was performed using a portable moisture probe, and at least five random samples from each batch were sent to the laboratory for calibration using the oven drying method (drying to constant weight at 105°C) to obtain the correction coefficient.

[0054] During the feeding period of the forage batch (defined as a continuous 14-day feeding phase), feeding data were collected from the target cattle herds consuming the forage batch. The data collected included feed intake (kg / day / herd or kg / day / cattle), feed intake rate (g / min / herd or g / min / cattle), rumination time (min / day / cattle), chewing frequency (times / minute / cattle), and average daily weight gain (ADG, kg / day).

[0055] Feed intake and feeding rate are automatically recorded by feeding devices or feed trough weighing devices equipped with weighing sensors, including the weight of feed before and after feeding and the weight of leftover feed, and are associated with timestamps and batch identifiers; rumination time and chewing frequency are extracted by a collar-type triaxial accelerometer worn by cattle and by a behavior recognition algorithm; average daily weight gain is measured and calculated by electronic scales at the beginning and end of each phase.

[0056] Align the forage quality records with the aquaculture data collected during the feeding period of that batch by batch identifier and timestamp to form a multimodal associated dataset. In the database, create an index with a unique batch identifier as the primary key for retrieval and version control.

[0057] Correlation and regression analyses were performed on the multimodal association dataset. Pearson correlation coefficient was used for correlation analysis, and linear regression was used for regression analysis. The quantitative relationship between moisture content and feed intake, feed rate, and ADG was evaluated. Time lag analysis was performed when necessary, such as examining the effect of moisture content on ADG in the last week or the last two weeks, to identify the delay effect. If the sample size is sufficient, a mixed-effects model can also be used to control for individual differences among cattle or random effects between batches.

[0058] Based on the above analysis results and according to the preset judgment rules, a comprehensive judgment is made on the forage drying assessment results, and the judgment results and correction suggestions are recorded in the database, including whether to mark the moisture content range as high quality, low quality or neutral, so as to be used for subsequent procurement, group feeding or processing target adjustment.

[0059] Specifically, this solution assigns a unique batch identifier to each batch of forage and links drying characteristic data such as moisture content, dry matter content, dryness grade, and drying uniformity to the collection time. During the feeding stage, breeding data such as feed intake, feeding rate, rumination time, chewing frequency, picky eating behavior, and average daily weight gain are all recorded with timestamps and mapped to the same batch identifier, achieving precise alignment of different data sources in the batch and time dimensions. This breaks through the single-indicator evaluation based solely on moisture content or dry matter, adding feeding behavior, feed intake, and production performance as evaluation dimensions. Furthermore, palatability evaluation results can be constructed and confidence or uncertainty indicators can be output, enabling the forage drying assessment to directly reflect the cattle herd's true acceptance and growth response to the batch of forage.

[0060] Using historical baselines as a reference, correlation or regression analysis and significance tests are performed on the feed intake rate and ADG changes in different moisture content ranges. Combined with health abnormality thresholds, an executable rule for judging high quality or poor quality is formed. When a certain dryness characteristic range increases feed intake and weight gain without increasing health risks, the system automatically marks it as a high quality dryness standard and updates it in a version, avoiding excessive drying and energy waste caused by fixed standards.

[0061] When the sample size is sufficient, mixed-effects models can be used to control for individual differences in cattle or random effects between batches, and time lag analysis can be supported to identify the delayed effects of drying conditions on production performance, thereby improving the robustness and interpretability of the evaluation conclusions.

[0062] A portable probe and oven calibration mechanism is introduced for moisture content detection. The average and standard deviation of moisture content in the same batch are statistically analyzed through multi-point sampling to form a drying uniformity index. The data storage module uses the batch identifier as the primary key to save the original data, related data, evaluation conclusions and standard update records, and supports version control. This facilitates the traceability of the basis for each evaluation and the source of standard changes, supporting the long-term operation and continuous optimization of large-scale ranches. The dynamically corrected drying evaluation standard can be configured hierarchically according to cattle herd type, breeding stage and forage species, and distributed to the ranch management system or terminal interface through the decision output interface. This is used to guide group feeding, forage priority ranking and processing target optimization, realizing the direct implementation from evaluation to production decision-making.

[0063] In one embodiment, the forage dryness grade is determined based on the moisture content / dry matter content of the forage batch, and the grading is determined using a preset threshold. Optionally, the moisture content is obtained by moisture measurement, and the dry matter content is a complementary amount to the moisture content, for example:

[0064] For hay, the average moisture content of each batch can be classified as follows: moisture content > 18% is considered slightly wet, 12% ≤ moisture content ≤ 18% is considered moderate, and moisture content < 12% is considered slightly dry.

[0065] For silage, the average dry matter content of a batch can be classified as follows: dry matter <30% is considered slightly moist, 30% ≤ dry matter ≤ 40% is considered moderate, and dry matter >40% is considered slightly dry.

[0066] It should be noted that the above thresholds are exemplary thresholds. In actual applications, they can be configured or dynamically modified according to grass species, regional climate, harvest season and aquaculture species, but the grading logic remains consistent.

[0067] Optionally, to characterize the drying uniformity of the same batch of forage, a uniformity index is calculated based on the moisture content measurements at multiple points within the same batch. For example, the uniformity index can be the standard deviation of moisture content σ or the coefficient of variation CV = σ / μ, where μ is the average moisture content of the batch; when CV is greater than a preset threshold (e.g., CV > 0.15), the batch can be marked as a batch with uneven drying risk.

[0068] In one embodiment, the forage drying assessment results include palatability evaluation results and confidence or uncertainty indicators. Optionally, the palatability evaluation results are output in the form of grades, including at least one or more of excellent, good, average, and poor.

[0069] For example, the rate of change in feed intake ΔFI, the rate of change in feed intake rate ΔFR, and the rate of change in production performance ΔP can be used as core response indicators, and the abnormal proportion A (such as the proportion of refusal to eat / picky eating behavior or the proportion of abnormal feeding events) can be used as a penalty indicator. When ΔFI ≥ 10% and ΔP ≥ 5% and A ≤ 5%, the output is excellent; when ΔFI ≥ 5% and ΔP ≥ 0 and A ≤ 8%, the output is good; when ΔFI is in [-5%, 5%] and A ≤ 10%, the output is average; when ΔFI < -5% or A > 10%, the output is poor.

[0070] Optionally, palatability evaluation results are output in the form of a score. For example, the overall score S can be calculated as follows: S = w1·ΔFI + w2·ΔFR + w3·ΔP − w4·A, where w1, w2, w3, and w4 are preset weight coefficients; the weight coefficients can be fitted based on historical data or configured by the manager; where w1 corresponds to the weight of the rate of change in feed intake, w2 corresponds to the weight of the rate of change in feed intake rate, w3 corresponds to the weight of the rate of change in production performance, and w4 corresponds to the weight of the penalty for abnormal proportions.

[0071] Furthermore, ΔFI represents the rate of change of feed intake relative to the historical baseline, ΔFR represents the rate of change of feed intake rate relative to the historical baseline, ΔP represents the rate of change of production performance relative to the historical baseline, and A represents the percentage of abnormal feed intake events and / or the percentage of abnormal cattle within the preset statistical period.

[0072] Furthermore, ΔFI = (current batch intake - baseline intake) / baseline intake × 100%, ΔFR = (current batch intake rate - baseline intake rate) / baseline intake rate × 100%, ΔP = (current batch ADG - baseline ADG) / baseline ADG × 100%; the baseline mean is taken from 30 days of historical data for the same grass species.

[0073] In one embodiment, confidence or uncertainty metrics are used to characterize the reliability of the evaluation results. For example, sample size, missing proportion, model goodness of fit, statistical significance test results, and prediction interval width can be used as factors in calculating confidence. When the statistical test shows a significant difference and the missing proportion is less than 10%, a high confidence level is output; when the test result is within the critical range or the missing proportion is between 10% and 30%, a medium confidence level is output; and when the missing proportion is greater than 30%, a low confidence level is output.

[0074] In one embodiment, in step S1, the forage quality profile further includes forage species information, source plot information, and sampling point information, which are used to characterize the consistency of drying status within the same forage batch.

[0075] In this embodiment, in addition to data such as moisture content and dry matter, the forage quality file also records the following information: grass species (e.g., whole corn plant, ryegrass, oat grass, or mixed grass), harvesting date, source plot number, harvesting and baling operation information, and sampling point information. At least 10 sampling points are randomly selected from each batch in the sampling point information, and the GPS coordinates and sampling time of each sampling point are recorded.

[0076] The drying uniformity index is obtained by calculating the mean and standard deviation of the moisture content of the same batch of samples. All of the above information is written into the batch file as structured fields, which supports subsequent stratified analysis by grass species or source plot.

[0077] In one embodiment, in step S2, the feed intake data includes the total feed intake of the forage batch within a preset time period and the feed intake rate per unit time. The feed intake rate is acquired in real time by a feeding device or feed trough weighing device equipped with a weighing sensor and mapped to the batch identifier.

[0078] In this embodiment, feed intake data includes two levels: the group level and the individual level. The total feed intake at the group level is statistically analyzed over a feeding cycle (14 days) as follows:

[0079]

[0080] Where d is the day number of the feeding stage, d=1,2,…,14.

[0081] The feeding rate is calculated in seconds or minutes from the weight change recorded by the weighing sensor in the feed trough and the feeding duration.

[0082]

[0083] The feeding duration refers to the length of time (in minutes) from the start to the end of a single feeding, which is calculated from the difference in timestamps recorded by the weighing sensor.

[0084] The weighing sensor uses a commercial feed trough weighing module (resolution ≤50g, sampling frequency 1Hz) and is mapped to the feed batch identifier API to ensure that each feeding data is written to the corresponding unique batch identifier.

[0085] In one embodiment, in step S2, the feeding behavior data includes at least one of the following: the herd's feeding activity level, rumination time, chewing frequency, and picky eating behavior characteristics, wherein at least part of the feeding behavior data is acquired by a collar-type behavior sensor and / or a camera image recognition device worn on the cattle.

[0086] In this embodiment, the feeding behavior data preferably includes rumination time, chewing frequency, and picky eating behavior indicators.

[0087] The specific implementation is as follows: Each cow is equipped with a collar with a three-axis accelerometer. The acceleration time-series signal is analyzed into behavioral states such as "feeding", "rumination" and "rest" through a trained behavior recognition algorithm, and the daily rumination duration (minutes / day) and average chewing frequency (times / minute) are counted.

[0088] Meanwhile, cameras are installed at several feeding trough locations, and image recognition algorithms are used to detect picky eating behavior (e.g., cattle pushing stems and discarding leaves with their noses, frequently raising their heads, etc.). The number of times or the duration of picky eating is used as the characteristics of picky eating behavior. This type of behavior data is aligned with the timestamp and unique batch identifier and entered into the multimodal association dataset.

[0089] In one embodiment, the construction of the multimodal association dataset employs an alignment rule based on batch identifiers and timestamps to associate forage quality records, feed intake, feeding behavior, and production performance data. For example, the feeding start time of a batch of forage is denoted as t0, and the feeding statistics window for that batch is defined as [t0, t0+24h]. Data such as feed intake, feeding rate, and feeding behavior collected within this window are attributed to that batch. For production performance data, a lag window can be used for association; for example, the change in production performance within [t0+24h, t0+72h] can be used as the lag response variable for that batch.

[0090] Optionally, when multiple batches of forage are fed together in the same time period, a weighted attribution method is used for association. For example, if the mass proportion of batch i in the feeding is denoted as wi (Σwi=1), then indicators such as feed intake and feeding rate can be weighted and assigned according to wi, or the feeding can be recorded as multiple batches of mixed feeding samples, and mixed feeding label features can be introduced into the model; where i represents the i-th batch of forage in the mixed feeding, wi represents the proportion of the i-th batch in the total mass of the mixed feeding, and Σwi=1 means that the sum of the mass proportions of each mixed feeding batch is 1.

[0091] Optionally, when there is cross-window impact from leftover material, a leftover material inheritance rule can be adopted. For example, if the proportion of leftover material to feed amount at the end of the previous window is greater than a preset threshold, it is marked as significant leftover material, and the batch information corresponding to the leftover material is inherited to the next window. The batch crosstalk flag is recorded in the associated dataset for subsequent weighting or removal in the analysis stage.

[0092] In this embodiment, the identification of cattle feeding behavior can be achieved using behavior recognition algorithms known in the art, which serve as a means of acquiring feeding behavior data.

[0093] For example, based on triaxial acceleration data collected by accelerometers worn on cattle, existing time-window-based behavior classification methods can be used to identify cattle's feeding, rumination, and resting behaviors.

[0094] Behavior classification methods may include, but are not limited to, any of the following:

[0095] (1) Behavior recognition methods based on statistical features and traditional classifiers, such as extracting features such as mean, variance, and spectral energy from acceleration data and then using support vector machines or random forests for classification;

[0096] (2) Neural network behavior recognition method based on time-series signals, such as using convolutional neural networks, recurrent neural networks or long short-term memory networks to perform end-to-end behavior classification of acceleration time-series data;

[0097] (3) Behavior recognition methods based on rules or template matching, such as frequency domain analysis of acceleration signals based on the periodic chewing characteristics of rumination behavior and recognition of rumination behavior by threshold determination.

[0098] The aforementioned behavior recognition algorithms are all well-known technologies in this field. This invention does not limit their specific implementation methods. As long as they can output data reflecting the behavior category, duration, or frequency of cattle feeding behavior, they can be used as the source of feeding behavior data in this invention.

[0099] In another embodiment, feeding behavior data can also be acquired through an image acquisition device, such as setting up a camera in the feeding area, and using target detection and behavior recognition algorithms known in the art to analyze the cattle feeding process.

[0100] For example, a target detection algorithm, such as a target detection algorithm based on a convolutional neural network, can be used first to locate the cattle, and then a time-series classification algorithm can be used based on a continuous image sequence to identify the cattle's feeding, picky eating, or rumination behavior.

[0101] The image recognition method described above is also existing technology, and its specific algorithm form does not constitute a limitation of this invention.

[0102] In one embodiment, in step S2, the production performance data includes the average daily weight gain (ADG) of the herd during the corresponding forage feeding phase, and may further include health status indicators to characterize adverse reactions.

[0103] In this embodiment, production performance data is mainly based on average daily weight gain (ADG). ADG is calculated by measuring the weight of the entire herd or individual cattle at the beginning and end of each phase using electronic scales. The ADG calculation is as follows:

[0104]

[0105] Health status indicators are recorded daily by the farm's veterinarian or feeder and stored in the database as structured event records (event type, cattle number, time, description). Digestive abnormalities (diarrhea, bloating, refusal to eat) are recorded in particular, and the abnormality incidence rate (number of abnormal cattle / total number of cattle) or abnormality incidence ratio (%) is used for subsequent judgment.

[0106] In one embodiment, step S4 involves analyzing the multimodal association dataset, including correlation and / or regression analysis between forage drying characteristics and feed intake, feeding behavior, and production performance, to determine the differences in breeding performance corresponding to different drying characteristic intervals.

[0107] Specifically, for the formed multimodal association dataset, descriptive statistics are first performed on continuous variables, including water content, foraging rate, ADG, etc., and descriptive statistics include mean, standard deviation and confidence interval.

[0108] Pearson correlation coefficient was used to assess the linear correlation between moisture content and foraging rate, and between moisture content and ADG; if the data did not meet the normal distribution, Spearman rank correlation was used.

[0109] Establish a linear regression model to quantify the contribution of each variable; when there are multiple repeated measures at the group / individual level, use a linear mixed-effects model to control for individual differences.

[0110] For the time lag effect, lag regression was used to test the delay effect, and the reliability of each association was assessed based on the statistical significance test results.

[0111] In one embodiment, step S5, the dynamic correction of the forage drying assessment criteria includes:

[0112] Based on the forage drying characteristics of increased feed intake, improved production performance, and no abnormal health reactions in cattle, the corresponding drying characteristics are marked as high-quality drying standards.

[0113] Based on the forage drying characteristics of reduced feed intake or decreased production performance in cattle, the corresponding drying characteristics are marked as inferior drying standards.

[0114] In this embodiment, the marking of high-quality drying standards and low-quality drying standards adopts the following specific judgment process. It should be noted that: for ease of implementation, this embodiment uses exemplary values, which can be adjusted through experiments according to the actual situation of the site.

[0115] Judgment process:

[0116] Baseline establishment: The average foraging rate and average ADG of the same grass species in the past 30 days in the field area were used as the historical baseline;

[0117] Judgment of increased feeding rate: If the average feeding rate in a certain moisture content range during the feeding stage increases by ≥10% compared to the baseline, and the difference is statistically significant after a statistical significance test, then the feeding rate is judged to have increased significantly.

[0118] Determination of ADG increase: If the ADG in the moisture content range increases by ≥5% compared to the baseline, and the statistical significance test is significant, then it is determined that ADG has increased significantly.

[0119] Health abnormality threshold: During the feeding period, if the incidence of digestive abnormalities is ≤5% and there is no significant increase from the baseline, it is considered that no abnormal health reaction has occurred;

[0120] Quality standard determination: When a significant increase in feeding rate and a significant increase in ADG occur simultaneously, and no abnormal health reactions are observed, the corresponding moisture content range is marked as a quality drying standard.

[0121] Determination of poor quality standards: If the rate of feed intake decreases significantly or the ADG decreases significantly, or the rate of digestive abnormalities increases significantly (exceeding the preset threshold), the corresponding moisture content range will be marked as poor quality drying standard.

[0122] The judgment results, statistical test reports, and raw data are all stored in the data storage module, and versioned standard update records are generated in the standard correction module.

[0123] In one embodiment, the dynamically modified forage drying assessment criteria are differentiated based on herd type, forage species, or stage of rearing to guide the optimization of group feeding, forage prioritization, or forage processing objectives.

[0124] Specifically, the dynamically revised drying assessment criteria are configured differently based on cattle herd type (such as fattening cattle, cows, and calves), forage species, and breeding stage (early / mid / late fattening).

[0125] For example, the system can record that the best performance of fattening cattle in the late stage is in the range of 15-17% moisture content, and then set priority for the use of forage with a moisture content of 15-17% for the fattening group; for cows, a higher moisture content priority range can be set to ensure the nutritional needs of lactation or pregnancy. The high-quality standard is mapped to different groups through group tags, and group feeding suggestions and procurement priorities are provided in the decision output interface.

[0126] Example 2: This example uses a dairy farm as an example. Whole-plant corn silage and partial silage were fed in TMR (Total Mixed Ration) form. The relationship between dryness grade and palatability or milk production performance was evaluated using the method in Example 1.

[0127] Each truckload of forage entering the TMR mixer is assigned a unique batch identifier ID, and a forage quality file is established. For each batch of forage, a sample is taken, and the moisture content at each point is obtained by moisture measurement. The average moisture content, dry matter content, and drying uniformity index are calculated. The batch is then classified into a drying grade according to a preset threshold. The above fields are bound to the timestamp and batch identifier and written into the file.

[0128] During this batch of TMR feeding, feed intake data and feeding behavior data are collected. Feed intake data is calculated by weighing the feed added and the leftover feed. Feeding behavior data is obtained by collar-type behavior sensors to obtain rumination time, chewing frequency, etc. A camera can be installed in the feeding trough to identify picky eating behavior and output the number of times or the duration of picky eating. Milk production in the same period is collected to reflect production performance response.

[0129] By aligning and associating forage records with feed intake, behavior, milk production, etc., based on batch identifiers and timestamps, a multimodal association dataset is formed. If there is cross-day residual feed, the residual feed can be attributed to the previous batch and the association rules are recorded.

[0130] Regression or correlation and statistical significance tests were performed on the associated datasets, with feed intake, rumination time, picky eating frequency, and milk yield as response variables; average moisture content, dryness grade, and uniformity index as independent variables; palatability evaluation results and confidence / uncertainty indexes corresponding to different dryness grades were output; based on this, it can be concluded that batches with moderate moisture content and good uniformity correspond to higher feed intake and more stable milk production, which can be used to guide subsequent dryness control targets.

[0131] Example 3: See Figures 1-6 A forage drying assessment system based on multimodal fusion, comprising:

[0132] The forage data acquisition module is used to collect drying characteristic data of forage batches and generate forage quality profiles;

[0133] The forage data acquisition module is implemented through the collaboration of multiple functional units, including a data acquisition unit, a data preprocessing unit, and a record generation unit;

[0134] The data acquisition unit is used to detect the drying characteristics of forage during the forage entry or storage stage, specifically including multi-point measurement of the moisture content of forage samples; the data preprocessing unit is used to process the collected raw test data, including outlier removal, averaging, or standardization; the archive generation unit is used to generate forage quality archives associated with forage batch identification and collection time based on the processed drying characteristic data.

[0135] Each batch of forage entering the facility is assigned a unique batch identifier upon arrival. The collected drying characteristic data includes the average moisture content, dry matter content, and a drying grade based on preset standards. The moisture content is measured at multiple sampling points using a portable moisture meter, and the average value is taken as the representative value for that batch. The forage quality file also records the collection time, the arrival time, and the corresponding batch identifier, providing a basis for subsequent data correlation.

[0136] The livestock data acquisition module is used to collect feed intake data, feeding behavior data, and production performance data of the corresponding cattle herd during batch feeding of forage;

[0137] The aquaculture data acquisition module consists of a feed intake acquisition unit, a feed behavior acquisition unit, and a production performance acquisition unit.

[0138] The feed intake acquisition unit is used to collect the amount of forage fed and the amount of forage remaining, and calculate the feed intake of the corresponding forage batch based on the difference between the two; the feeding behavior acquisition unit is used to collect behavioral data of cattle during feeding, including acceleration signals obtained by behavior sensors or behavioral images obtained by image acquisition devices; the production performance acquisition unit is used to collect weight change data of cattle in the corresponding feeding stage, and is used to calculate the average daily weight gain index.

[0139] Feed intake data is obtained through feeding devices installed in the feeding area. The feeding devices have built-in weighing sensors to record the amount of feed given and the amount of feed left over in each feeding cycle, and to calculate the actual feed intake of the corresponding forage batch. Feeding behavior data is obtained through collar-type behavior sensors worn around the necks of cattle. The behavior sensors collect the acceleration signals of the cattle during feeding, and combine them with existing behavior recognition algorithms to convert the collected signals into feeding behavior categories and corresponding durations or frequencies. This behavior recognition process is a known technical means in the field and is only used to obtain feeding behavior data. Production performance data in this embodiment mainly includes the average daily weight gain of the herd during the feeding phase of the forage batch, which is obtained by weighing the herd at the beginning and end of the feeding phase.

[0140] The data association module is used to associate forage quality records with breeding data by batch and time based on timestamps and forage batch identifiers, and to build a multimodal association dataset.

[0141] The data association module includes a time alignment unit, a batch matching unit, and a data integration unit.

[0142] The time alignment unit is used to uniformly process the timestamps of data from different data sources to eliminate differences in collection frequency or collection start and end time; the batch matching unit is used to filter the breeding data corresponding to the current forage batch based on the forage batch identifier; and the data integration unit is used to combine the data that have completed time alignment and batch matching to form a multimodal association dataset containing forage drying characteristic data, feed intake data, feed behavior data and production performance data.

[0143] The system uses the forage batch identifier as the main index to match the forage intake data, forage behavior data and production performance data collected within the same time period with the corresponding forage quality files, forming a multimodal association dataset.

[0144] The data storage module is used to store raw data, processed data, and evaluation results indexed by batch identifiers and supports version control. The data storage module supports data version management based on batch identifiers so that the results of the same forage batch under different evaluation periods can be traced and compared.

[0145] The data storage module consists of a raw data storage unit, a related data storage unit, and a result storage unit.

[0146] The raw data storage unit is used to store unprocessed sensor data or detection data; the associated data storage unit is used to store multimodal associated datasets processed by the data association module; the result storage unit is used to store the evaluation results output by the analysis and evaluation module and the corrected drying evaluation criteria. The data storage module supports data version management based on forage batch identification to facilitate the traceability of historical data.

[0147] The analysis and evaluation module is used to analyze multimodal association datasets to obtain feeding performance and production performance response results corresponding to forage drying characteristics;

[0148] The standard correction module is used to dynamically update the forage drying assessment standard based on the output results of the analysis and evaluation module, and to send the updated results to the ranch management system or display interface through the decision output interface.

[0149] The standard correction module includes a rule judgment unit and a standard update unit.

[0150] The system includes a rule-based judgment unit to determine whether the current forage drying characteristics meet the preset criteria for high-quality or low-quality based on the analysis and evaluation results; and a standard update unit to adjust the forage drying evaluation criteria when the update criteria are met. The decision output interface includes a data interface unit and a display control unit, used to output the updated evaluation criteria to the ranch management system or display them on the user terminal.

[0151] The analysis results output by the analysis and evaluation module are transmitted as input to the standard correction module. The standard correction module updates the forage drying evaluation standard based on the analysis results. The updated evaluation results are sent to the ranch management system through the decision output interface provided by the system, or displayed in a visual manner on the management terminal interface to guide subsequent forage procurement, processing or feeding decisions.

[0152] In this embodiment, the analysis and evaluation module includes a multimodal fusion analysis unit. The multimodal fusion analysis unit is used to comprehensively utilize forage drying characteristic data, feed intake data, feed behavior data and production performance data, and adopts correlation analysis, regression analysis and statistical significance test procedures to comprehensively evaluate the forage drying status and output palatability evaluation results and confidence or uncertainty indicators.

[0153] In practical implementation, the input data received by the multimodal fusion analysis unit includes:

[0154] The data includes drying characteristics, feed intake, feeding behavior statistics, and production performance data corresponding to each batch of forage. Among them, the feeding behavior data is converted into structured indicators before entering the analysis process, such as the duration of feeding behavior, the duration of rumination behavior, or the frequency of feeding behavior.

[0155] The multimodal fusion analysis unit processes the above data according to a preset analysis process. This process includes a correlation analysis step to assess the correlation between forage drying characteristics and feed intake, feeding behavior, and production performance indicators; it also includes a regression analysis step to establish a mathematical relationship model between changes in forage drying characteristics and changes in production performance; to improve the reliability of the evaluation results, a statistical significance test step is further introduced into the analysis process to determine whether the obtained correlation or regression results are statistically significant. The above analysis methods are all conventional data analysis methods in this field, and their specific implementation methods can be selected according to actual application needs.

[0156] Based on the above analysis results, the multimodal fusion analysis unit outputs comprehensive evaluation results for different forage drying states. In this embodiment, the evaluation results include at least the forage palatability evaluation results, and may further include the confidence or uncertainty index of the corresponding evaluation results to reflect the stability or credibility of the evaluation results. The output results are the final output of the analysis and evaluation module to support the dynamic correction of subsequent forage drying evaluation standards.

[0157] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for assessing forage drying based on multimodal fusion, characterized in that, Includes the following steps: S1. Assign a unique batch identifier to each batch of forage and establish a forage quality file. Collect and record the drying characteristic data of the forage. The drying characteristic data includes, but is not limited to, moisture content, dry matter content, dryness grade and drying uniformity index, and associate it with the time information and batch identifier of the forage. S2. During the feeding of the forage batches, the target cattle herd is fed with a variety of feeding data, including at least two of the following: feed intake data, feeding behavior data, and production performance data. S3. Based on the batch identifier and timestamp, associate the forage quality file with the breeding data to construct a multimodal association dataset between forage drying characteristics and cattle breeding performance; S4. Analyze the multimodal association dataset to obtain the feeding performance and production performance response results corresponding to different forage drying characteristics; S5. Based on the feeding performance and production performance response results, make a comprehensive judgment on the forage drying assessment results, and dynamically revise the forage drying assessment standard accordingly, so that the revised forage drying assessment standard can reflect the actual feeding effect and production performance of the cattle herd.

2. The forage drying assessment method based on multimodal fusion according to claim 1, characterized in that, In step S1, the forage quality file further includes forage species information, source plot information, and sampling point information, which are used to characterize the consistency of drying status within the same forage batch.

3. The forage drying assessment method based on multimodal fusion according to claim 1, characterized in that, In step S2, the feed intake data includes the total feed intake of the forage batch within a preset time period and the feed intake rate per unit time. The feed intake rate is acquired in real time by a feeding device or feed trough weighing device equipped with a weighing sensor and mapped with the batch identifier.

4. The forage drying assessment method based on multimodal fusion according to claim 1, characterized in that, In step S2, the feeding behavior data includes at least one of the following: the feeding activity level of the herd, rumination time, chewing frequency, and picky eating behavior characteristics, wherein at least part of the feeding behavior data is acquired by collar-type behavior sensors and / or camera image recognition devices worn on the cattle.

5. The forage drying assessment method based on multimodal fusion according to claim 1, characterized in that, In step S2, the production performance data includes the average daily weight gain (ADG) of the herd during the corresponding forage feeding phase, and may further include health status indicators to characterize adverse reactions.

6. The forage drying assessment method based on multimodal fusion according to claim 1, characterized in that, In step S4, the analysis of the multimodal association dataset includes correlation analysis and / or regression analysis between forage drying characteristics and feed intake, feeding behavior and production performance, in order to determine the differences in breeding performance corresponding to different drying characteristic intervals.

7. The forage drying assessment method based on multimodal fusion according to claim 1, characterized in that, In step S5, the dynamic correction of the forage drying assessment criteria includes: Based on the forage drying characteristics of increased feed intake, improved production performance, and no abnormal health reactions in cattle, the corresponding drying characteristics are marked as high-quality drying standards. Based on the forage drying characteristics of reduced feed intake or decreased production performance in cattle, the corresponding drying characteristics are marked as inferior drying standards.

8. The forage drying assessment method based on multimodal fusion according to claim 7, characterized in that, The dynamically revised forage drying assessment criteria are set differently based on cattle herd type, forage species, or breeding stage to guide the optimization of group feeding, forage priority ranking, or forage processing targets.

9. A forage drying assessment system based on multimodal fusion, used to implement the forage drying assessment method based on multimodal fusion as described in any one of claims 1-8, characterized in that, The system includes: The forage data acquisition module is used to collect the drying characteristic data of the forage batch and generate forage quality files; The livestock data acquisition module is used to collect feed intake data, feeding behavior data, and production performance data of the corresponding cattle herd during the feeding of the forage batches. The data association module is used to associate the forage quality file with the breeding data by batch and time based on the timestamp and the forage batch identifier, and to construct a multimodal association dataset. The data storage module is used to store raw data, processed data, and evaluation results indexed by batch identifiers and supports version control. The analysis and evaluation module is used to analyze the multimodal association dataset to obtain the feeding performance and production performance response results corresponding to the forage drying characteristics; The standard correction module is used to dynamically update the forage drying assessment standard based on the output results of the analysis and evaluation module, and send the updated results to the ranch management system or display interface through the decision output interface.

10. The forage drying assessment system based on multimodal fusion according to claim 9, characterized in that, The analysis and evaluation module includes a multimodal fusion analysis unit, which is used to comprehensively utilize forage drying characteristic data, feed intake data, feed behavior data and production performance data, and adopt correlation analysis, regression analysis and statistical significance test procedures to comprehensively evaluate the forage drying status and output palatability evaluation results and confidence or uncertainty indicators.

Images