Plant growth prediction method and device based on growth state prediction large model, equipment and medium

Abstract

The invention relates to a plant growth prediction method and device based on a growth state prediction large model, equipment and a medium, and belongs to the technical field of plant prediction.The method comprises the steps that feature detection data of a target plant object and environment detection data of a facility growth environment where the target plant object is located are obtained through a detection device; inputting the feature detection data and the environment detection data into a preset growth state prediction large model to obtain a growth state prediction result of the target plant object; inputting the growth state prediction result into a preset evaluation model to obtain an evaluation result; and adjusting operation parameters of an environment adjusting device configured for the facility growth environment based on the evaluation result.

Description

Technical Field

[0001] This disclosure relates to the technical field of plant prediction, and more specifically, to a method, apparatus, device, and medium for plant growth prediction based on a large-scale growth state prediction model. Background Technology

[0002] Currently, tomatoes, as a high-value-added cash crop, occupy an important position in my country's large-scale and intensive agricultural production. Their yield and quality directly affect agricultural economic benefits and food security. Tomato production is characterized by a short growing season, a controllable environment but influenced by multiple coupled factors (temperature, humidity, light, CO2, soil moisture, etc.), and labor-intensive management. Existing facility agriculture monitoring systems mostly rely on multiple sensors to collect environmental data and use rule-based or simple statistical learning and machine learning models for threshold alarms or growth status prediction. However, these systems suffer from problems such as data silos, insufficient utilization of cross-modal data, and weak decision-making capabilities. Summary of the Invention

[0003] One objective of this disclosure is to provide a new technical solution for plant growth prediction based on a large-scale growth state prediction model.

[0004] According to a first aspect of this disclosure, a plant growth prediction method based on a large-scale growth state prediction model is provided, the method comprising: The detection device acquires feature detection data of the target plant object and environmental detection data of the growing environment of the facility where the target plant object is located. The feature detection data and the environmental detection data are input into a pre-set large growth state prediction model to obtain the growth state prediction result of the target plant object. The growth state prediction results are input into a preset evaluation model to obtain the evaluation results; Based on the assessment results, the operating parameters of the environmental control device configured for the growth environment of the facility are adjusted.

[0005] Optionally, the large growth state prediction model includes a first input layer, a feature fusion layer, and a prediction layer; The step of inputting the feature detection data and the environmental detection data into a pre-set large-scale growth state prediction model to obtain the growth state prediction result of the target plant object includes: The feature detection data and the environmental detection data are input into the first input layer to obtain the instruction text for the target plant object and the multimodal features of the facility's growth environment; The instruction text and the multimodal features are input into the feature fusion layer to obtain the embedding vector; The embedding vector is input into the prediction layer to obtain a binary classification probability, which is used as the prediction result of the growth status of the target plant object.

[0006] Optionally, the method further includes: Obtain the first training sample set; The large-scale growth state prediction model is trained using the first training sample set to obtain the trained large-scale growth state prediction model.

[0007] Optionally, after training the large-scale growth state prediction model using the first training sample set to obtain the trained large-scale growth state prediction model, the method further includes: The output of the growth state prediction model to the first training sample set and the first training sample are input into a preset verification algorithm to obtain training values. If the training values ​​exceed the set baseline value, the training of the large growth state prediction model is terminated.

[0008] Optionally, the evaluation model includes a second input layer, a retrieval layer, and a generation layer; The step of inputting the growth state prediction result into a preset evaluation model to obtain the evaluation result includes: The growth state prediction results are input into the second input layer to obtain the growth state label and the multimodal features of the facility's growth environment; The growth state label and the multimodal features are input into the retrieval layer to obtain the retrieval results; The search results are input into the generation layer to obtain the measure text, which serves as the evaluation result.

[0009] Optionally, the method further includes: Obtain the second training sample set; The evaluation model is trained using the second training sample set to obtain the trained evaluation model.

[0010] Optionally, the second training sample set includes a first training data set with real labels and a second training data set with pseudo-labels; The second training data set is constructed from the first training data set.

[0011] According to a second aspect of this disclosure, a plant growth prediction device based on a large-scale growth state prediction model is also provided, the device comprising: The acquisition module is used to acquire feature detection data of the target plant object and environmental detection data of the growth environment of the facility where the target plant object is located through the detection device; The first obtaining module is used to input the feature detection data and the environmental detection data into a preset growth state prediction model to obtain the growth state prediction result of the target plant object; The second obtaining module is used to input the growth state prediction results into a preset evaluation model to obtain the evaluation results; The adjustment module is used to adjust the operating parameters of the environmental regulation device configured for the growth environment of the facility based on the evaluation results.

[0012] According to a third aspect of this disclosure, a computer system is also provided, the computer system including a processor, which, when executing program instructions or code, implements the plant growth prediction method based on a large growth state prediction model in the first aspect.

[0013] For example, the computer system also includes a memory for storing program instructions or code.

[0014] According to a fourth aspect of this disclosure, a computer-readable storage medium is also provided, wherein a computer program is stored therein, wherein the computer program is configured to execute the above-described plant growth prediction method based on a large growth state prediction model at runtime.

[0015] According to a fifth aspect of this disclosure, a computer program product is also provided, comprising a computer program that, when executed, causes a computer to perform the steps of the above-described plant growth prediction method based on a large growth state prediction model.

[0016] According to a sixth aspect of this disclosure, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the above-described plant growth prediction method based on a large growth state prediction model through the computer program.

[0017] One beneficial effect of this disclosure is that the plant growth prediction method based on the large growth state prediction model provided by the present invention can determine whether the target plant is growing well in the facility growth environment through the configured large growth state prediction model and evaluation model. If the growth is poor, the facility growth environment can be adjusted by the environmental adjustment device configured in the facility growth environment, thereby effectively improving the growth efficiency of the target plant.

[0018] Other features and advantages of the embodiments of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments of the present disclosure and, together with their description, serve to explain the principles of the embodiments of the present disclosure.

[0020] Figure 1 A flowchart illustrating a plant growth prediction method based on a large growth state prediction model, according to some embodiments, is shown. Figure 2 A schematic diagram of a plant growth prediction device based on a large growth state prediction model according to some embodiments is shown. Figure 3 A schematic diagram of the hardware structure of an electronic device according to some embodiments is shown. Detailed Implementation

[0021] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0022] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0023] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0024] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0025] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0026] <Method Implementation> Figure 1 This is a flowchart illustrating a plant growth prediction method based on a large-scale growth state prediction model according to one embodiment. The implementing entity is a smart terminal, such as a personal computer, mobile phone, tablet, or server.

[0027] like Figure 1 As shown, the plant growth prediction method based on a large growth state prediction model in this embodiment may include the following steps S110 to S140: Step S110: Using a detection device, acquire feature detection data of the target plant object and environmental detection data of the growing environment of the facility where the target plant object is located.

[0028] In this embodiment, the target plant object can be a tomato seedling, and the feature detection data of the target plant object can be data collected by the sensor in the detection device. The feature detection data includes: stem diameter (unit: mm, accuracy 0.1 mm), plant height (unit: cm, accuracy 0.1 cm), leaf area (unit: cm², accuracy 1 cm²), number of leaves (unit: leaves, integer), and chlorophyll content (unit: SPAD value, range 0-100).

[0029] In this embodiment, the growing environment can be a greenhouse, and the environmental monitoring data can be obtained through multi-point environmental monitoring sensors in the monitoring device and an outdoor weather station. The environmental monitoring data here can include greenhouse environmental data, soil data, and outdoor meteorological data. Greenhouse environmental data includes air temperature (°C), air humidity (% RH), light intensity (lux), and CO2 concentration (ppm); soil data includes soil temperature (°C), soil moisture (% vol), soil electrical conductivity (mS / cm), and soil pH; outdoor meteorological data includes wind speed (m / s), rainfall (mm), dew point temperature (°C), and air pressure (hPa).

[0030] Step S120: Input the feature detection data and the environmental detection data into the preset growth state prediction model to obtain the growth state prediction result of the target plant object.

[0031] In some embodiments, the large-scale growth state prediction model includes a first input layer, a feature fusion layer, and a prediction layer; step S120 may include the following steps S210 to S230: Step S210: Input the feature detection data and the environmental detection data into the first input layer to obtain the instruction text for the target plant object and the multimodal features of the facility growth environment.

[0032] Step S220: Input the instruction text and the multimodal features into the feature fusion layer to obtain the embedding vector.

[0033] Step S230: Input the embedding vector into the prediction layer to obtain a binary classification probability, which is used as the prediction result of the growth status of the target plant object.

[0034] In this embodiment, the large-scale growth state prediction model can be based on a general large-scale model, incorporating knowledge of the tomato domain to construct the "Qwen3-7B-Tomato" base model.

[0035] In this embodiment, the domain text library configured for the large-scale growth state prediction model is obtained by integrating text data (such as professional materials, agricultural records, and question-and-answer pairs), removing duplicate content, and filtering invalid text. General text supplement: A 1:1 ratio of Wikipedia Chinese general text is provided to prevent the large-scale growth state prediction model from losing its general question-answering capabilities during domain pre-training, mitigating the "disaster forgetting" phenomenon. Furthermore, the initial weights of this large-scale growth state prediction model can be the weights of the Qwen3-8B general-purpose large-scale model, retaining its Transformer basic architecture (including RoPE positional encoding, SwiGLU activation function, RMSnorm normalization, and Attention QKV bias). Hyperparameter configuration: Optimal hyperparameters are determined based on the dataset size and hardware computing power, as follows: cutoff_len = 1024, per_device_train_batch_size = 2, gradient_accumulation_steps = 8, learning_rate = 1.00E-04, num_train_epochs = 5, lr_scheduler_type = cosine, warmup_ratio = 0.1, bf16 = TRUE, and weight_decay = 0.01. The hardware and software environment used in the large growth state prediction model can be: Nvidia A100 GPU (40GB VRAM), CUDA 12.6, PyTorch 2.3.1, Transformers 4.38.2.

[0036] In this embodiment, the input layer can receive both feature detection data and environmental detection data, and fuse them into a unified input using JSON format. The instruction text for the target plant object output by the input layer is, for example, "Based on n days of historical greenhouse indoor and outdoor environmental monitoring data, predict the tomato seedling growth status label (good growth / poor growth). Missing data has been filled with 0; please handle it appropriately." The multimodal features of the target plant object output by the input layer are, for example, a sequence of numerical data (environment, soil, meteorology) from time tn to time t. The feature fusion layer can employ a "numerical embedding + text embedding" fusion strategy, converting the instruction text and multimodal features into embedding vectors using the Qwen3-7B-Tomato embedding layer. The prediction layer can consist of an RMSnorm regularization layer, a fully connected layer, and a Softmax layer, outputting a binary classification probability as the predicted growth status of the target plant object.

[0037] In some embodiments, the method further includes the following steps S310 and S320: Step S310: Obtain the first training sample set.

[0038] Step S320: Train the large growth state prediction model using the first training sample set to obtain the trained large growth state prediction model.

[0039] In this embodiment, the first training sample set needs to label the training samples in the first training sample set with growth status labels to determine the binary classification labels of the training samples ("good growth" and "poor growth"). Referring to the seedling strength index calculation formula (seedling strength index = stem diameter / plant height × total plant dry weight, and for simplified calculation, "stem diameter ≥ 8mm and plant height 10-15cm and number of leaves ≥ 6" is used as the threshold for "good growth"), the growth status labels of the target plant objects at each measurement time point are labeled.

[0040] In this embodiment, the multimodal features of the training samples may include data such as stem diameter, plant height, and leaf area of ​​the target plant object, and be labeled with measurement time, plant number, and growth stage (seedling stage, flowering and fruit setting stage, etc.).

[0041] In this embodiment, the instruction text of the training samples may include joint annotation using "human + BERT model" to extract two types of key knowledge: growth judgment rules, such as "light intensity < 10000 lux for 2 consecutive days → growth may be poor" and "daily plant height increase < 0.3 cm → poor growth"; and measure suggestion knowledge, such as "poor growth and soil moisture < 30% → increase irrigation to soil moisture 60%-70%" and "night temperature < 15℃ → turn on heating equipment to 18℃ for 4 hours".

[0042] In this embodiment, large-scale automatic data labeling is achieved by employing "pseudo-label learning + data augmentation." A lightweight classification growth state prediction model is trained with precisely labeled data, pseudo-labels are generated for unlabeled data, and the dataset is expanded through "numerical perturbation" (e.g., adding ±0.5℃ noise to temperature data) and "text synonym replacement" (e.g., replacing "increase illumination" with "increase illumination intensity"). A 10% sample of the automatically labeled data is manually proofread to ensure label accuracy ≥90%, ultimately yielding the first training sample set (containing numerical values, text, ontology data, corresponding growth state labels, and measure knowledge).

[0043] In this embodiment, the training process of the large-scale growth status prediction model is as follows: A self-supervised learning approach is adopted, and pre-training is performed on a domain text library and general texts. The training objectives include: Model naming: After training, the tomato domain base model is obtained and named "Qwen3-7B-Tomato"; Validation method: The model's domain capability is verified through the "Tomato Pest and Disease Entity and Relationship Triple Joint Extraction Task". Input text containing descriptions of pests and diseases (e.g., "Yellow spots appear on tomato leaves, gradually expanding to the entire leaf"), and the large-scale growth status prediction model needs to extract triples, such as <tomato leaves, symptoms, yellow spots>, <yellow spots, cause, late blight>, etc.; Validation metrics: Precision, Recall, and Harmonic Mean (F1) reach or exceed the current benchmark, ensuring the model has the ability to extract domain knowledge. Based on the Qwen3-7B-Tomato base model, a large-scale growth status prediction model with a Transformer architecture is built to achieve binary classification prediction of "good growth / poor growth" for plant objects in the seedling stage.

[0044] In this embodiment, the input format of the training samples in the first training sample set adopts the JSON structured format, as follows: { "Window start date": "2024-03-01 08:00:00", "Window End Date": "2024-03-07 08:00:00", "Growth Days": "21 days (21 days after sowing)" "Environmental data sequence": { Temperature: [24.53, 25.25, 27.33, ..., 26.81] Humidity: [75.38, 75.98, 73.75, ..., 74.22], "Illumination": [18500, 22300, 0, ..., 21800], "CO2": [850, 880, 920, ..., 860], Soil temperature: [22.1, 22.5, 23.0, ..., 22.7], Soil moisture: [65.2, 64.8, 63.5, ..., 65.0], Soil electrical conductivity: [2.1, 2.0, 2.2, ..., 2.1], Soil pH: [6.5, 6.5, 6.6, ..., 6.5], Wind speed: [1.2, 0.8, 1.5, ..., 1.0], Rainfall: [0, 0, 5.2, ..., 0], "Dew point temperature": [18.2, 18.5, 19.0, ..., 18.7], "Atmospheric pressure": [1012, 1013, 1011, ..., 1012] }, } Output format: Binary labels in JSON format, as shown in the example below: { "Growth Forecast Result": "Growth is good" } In this embodiment, the initial weights of the large-scale growth state prediction model can be the weights of the Qwen3-7B-Tomato pedestal model. Loss function: The cross-entropy loss function is used, with the following specific expression: Lce = N 1∑ i =1 N [ yi log( pi )+(1 yi log(1) pi )] in, N For the sample size, yi For the first i The true labels for each sample ("good growth" = 1, "poor growth" = 0). pi To predict the probability of "good growth" for the model.

[0045] Among them, the optimizer is AdamW optimizer. β 1 = 0.9 β 2=0.999, ε=1e-8), combined with gradient clipping (maximum gradient norm=1.0) to prevent gradient explosion.

[0046] In this embodiment, a large-scale growth state prediction model is trained using training samples of "air environment data," "soil data + air environment data," and "full data (air + soil + meteorological + ontology)," and the data combination with the best performance is selected. Sampling frequency comparison: sampling frequencies of "1 time / hour," "1 time / 4 hours," and "1 time / day" are used to determine the optimal time resolution; sequence length comparison: historical data windows of "3 days," "5 days," and "7 days" are used to select the optimal input sequence length.

[0047] In this embodiment, the initial weights of the large growth state prediction model are loaded, and all Transformer layers are pre-trained. During the training process, the weights of the large growth state prediction model are saved every 1000 steps, and the weights with the smallest loss on the validation set are selected as the final weights of the large growth state prediction model (Qwen3-7B-Tomato).

[0048] In some embodiments, after step S320, the method further includes the following steps S410 and S420: Step S410: Input the output of the growth state prediction model to the first training sample set and the first training sample into a preset verification algorithm to obtain training values.

[0049] Step S420: If the training value exceeds the set benchmark value, the model training of the large growth state prediction model is terminated.

[0050] In this embodiment, the expression for the verification algorithm is as follows: Where Accuracy represents classification accuracy, F1 represents the harmonic mean, TP (true positive) is the number of samples that are both "actually good and predicted to be good", TN (true negative) is the number of samples that are both "actually poor and predicted to be poor", FP (false positive) is the number of samples that are "actually poor but predicted to be good", and FN (false negative) is the number of samples that are "actually good but predicted to be poor". If the accuracy of the large-scale growth status prediction model on the first training sample set reaches or exceeds the set benchmark value, the prediction accuracy of the large-scale growth status prediction model is determined to meet the actual production requirements. Here, the accuracy is the training value.

[0051] Step S130: Input the growth state prediction result into the preset evaluation model to obtain the evaluation result.

[0052] In some embodiments, the evaluation model includes a second input layer, a retrieval layer, and a generation layer; step S130 may include the following steps S510 to S530: Step S510: Input the growth state prediction result into the second input layer to obtain the growth state label and the multimodal features of the facility growth environment.

[0053] Step S520: Input the growth state label and the multimodal features into the retrieval layer to obtain the retrieval results.

[0054] Step S530: Input the search results into the generation layer to obtain the measure text, which is used as the evaluation result.

[0055] In this embodiment, the second input layer receives the growth status prediction result, multimodal features, expert knowledge set, and instruction text, which are then fused using JSON format. The growth status prediction result is output as either a "good growth" or "poor growth" label. The expert knowledge set may include growth judgment rules and recommended measures. The instruction text may include targeted measures generated based on the input growth status label, multimodal features, and expert knowledge set. Specifically, if the plant's growth status is "good growth," consolidation measures are output; if the plant's growth status is "poor growth," remedial measures (including specific parameters such as temperature and light intensity) are output. In this embodiment, the retrieval layer is a knowledge retrieval layer, which adopts retrieval-enhanced generation (RAG) technology. For the input multimodal features, it retrieves the three most relevant expert knowledge from the pre-set measure library (e.g., input "soil moisture 30% + growth difference", retrieve "growth difference and soil moisture < 30% → increase irrigation amount to 60%-70%), as knowledge support for the evaluation model generation.

[0056] In this embodiment, the generation layer is based on a Transformer decoder, combining the domain knowledge of the base model with the retrieved expert knowledge to generate measure text as the evaluation result. The output format of the measure text is structured natural language (listed in bullet points, including specific parameters).

[0057] In some embodiments, the method further includes the following steps S610 and S620: Step S610: Obtain the second training sample set.

[0058] Step S620: Train the evaluation model using the second training sample set to obtain the trained evaluation model.

[0059] In this embodiment, the initial weights of the evaluation model are the model weights loaded with the evaluation model, and all Transformer decoder parameters are fine-tuned.

[0060] In this embodiment, the loss function of the evaluation model is the BLEU loss function (based on n-gram similarity). Taking n=4 as an example, the specific expression of the loss function is as follows: in, For the reference text (expert measures) length, To evaluate the length of the text generated by the model, It is n-gram precision.

[0061] In this embodiment, by using "gradient accumulation steps=4" and "learning rate warm-up steps=1000", loss oscillations in the early stage of model training are avoided, and "length constraints" are applied to the generated text to ensure that the content of the output measure text is complete and concise.

[0062] In some embodiments, the second training sample set includes a first training data set with real labels and a second training data set with pseudo labels; The second training data set is constructed from the first training data set.

[0063] In this embodiment, the first training dataset can be cross-modal features (including growth status labels, environmental data, and plant data) and expert intervention text, augmented using a "teacher-student" distillation method. Using an existing closed-source large model, inputting "environmental data + plant data + growth status labels," and issuing instructions (such as "generate targeted measures for tomato seedlings, including specific parameters"), it generates high-quality intervention text, thus obtaining the first training dataset for the teacher model. The first training dataset is then paired with manually annotated expert intervention text to construct "input-output" training data. The input format of this training data is JSON structured format, as shown in the example below ("good growth" scenario): { "Window start date": "2024-03-01 08:00:00", "Window End Date": "2024-03-07 08:00:00", "Growth days": "21 days" "Environmental data sequence": { Temperature: [24.53, 25.25, ..., 26.81] Humidity: [75.38, 75.98, ..., 74.22], "Illumination": [18500, 22300, ..., 21800], "...": "..." }, "Plant intrinsic data": { Stem diameter: 8.2mm Plant height: 12.5cm Leaf area: 14.3 cm² Number of leaves: 7 Chlorophyll: 52 SPAD}, Growth Status Tag: "Growing Well" Expert recommendations: "When growth is good, maintain 18,000-22,000 lux of sunlight, 60%-70% soil moisture, and a diurnal temperature range of 8-10℃." } The output format of the training data is JSON, and the corresponding measure text example is as follows: { Recommended countermeasures: - Strengthen light management: Maintain a daily light intensity of 18,000-22,000 lux for 12-14 hours, and turn on supplemental lighting (200W / ㎡) if necessary; - Optimize soil moisture: Maintain soil moisture at 60%-70%, using drip irrigation, with an irrigation amount of 50-100ml / plant each time, every 2 days; - Control diurnal temperature difference: Daytime temperature 25-28℃, nighttime temperature 18-20℃, maintain a temperature difference of 8-10℃, and avoid excessively high nighttime temperatures that may cause excessive vegetative growth. } In this embodiment, DeepSeek-R1 is used as the teacher model. The input is "multimodal features + growth status label," and the command is "generate targeted measures for the plant object, including specific parameters (such as temperature and concentration), listed in bullet points, 150 characters." Example input: "Environmental data: temperature 22-24℃, humidity 75%-80%, light 15000-18000 lux, soil moisture 65%; Plant data: stem diameter 8.1mm, plant height 12.3cm, number of leaves 7, chlorophyll 51; Growth status label: healthy growth." The teacher model outputs: "1. Light Management: Maintain a daily light intensity of 15,000-18,000 lux for 12-14 hours. If it is cloudy, turn on supplemental lighting (200W / ㎡) to ensure sufficient light; 2. Soil Moisture: Maintain soil moisture at 60%-70%, using drip irrigation, 80ml per plant per irrigation, every 2 days; 3. Temperature Control: Daytime temperature 25-28℃, nighttime temperature 18-20℃, diurnal temperature difference 8-10℃, avoid excessive growth caused by high nighttime temperatures." The text generated by the teacher model is combined with the manually annotated expert measures text to construct the "input-output" training data, thus obtaining the second training sample set.

[0064] Step S140: Based on the evaluation results, adjust the operating parameters of the environmental control device configured for the facility's growth environment.

[0065] In this embodiment, the environmental control device is, for example, a humidifier or a temperature regulator installed in the facility's growth environment, and is not limited to this.

[0066] In this embodiment, by configuring a large-scale growth status prediction model and an evaluation model, it is determined whether the target plant is growing well in the facility growth environment. If the growth is poor, the facility growth environment can be adjusted by the environmental adjustment device configured in the facility growth environment, thereby effectively improving the growth efficiency of the target plant.

[0067] <Equipment Example 1> Figure 2 This is a schematic diagram of a plant growth prediction device based on a large-scale growth state prediction model, according to one embodiment. Figure 2 As shown, the plant growth prediction device 200 based on a large-scale growth state prediction model may include: The acquisition module 210 is used to acquire feature detection data of the target plant object and environmental detection data of the growth environment of the facility where the target plant object is located through the detection device; The first obtaining module 220 is used to input the feature detection data and the environmental detection data into a preset growth state prediction model to obtain the growth state prediction result of the target plant object; The second module 230 is used to input the growth state prediction result into a preset evaluation model to obtain the evaluation result; The adjustment module 240 is used to adjust the operating parameters of the environmental control device configured for the growth environment of the facility based on the evaluation results.

[0068] Optionally, the first obtaining module 220 is further configured to input the feature detection data and the environment detection data into the first input layer to obtain the instruction text for the target plant object and the multimodal features of the facility growth environment; input the instruction text and the multimodal features into the feature fusion layer to obtain the embedding vector; and input the embedding vector into the prediction layer to obtain the binary classification probability, which is used as the prediction result of the growth state of the target plant object.

[0069] Optionally, the device further includes a first training module for acquiring a first training sample set; and training the large growth state prediction model using the first training sample set to obtain the trained large growth state prediction model.

[0070] Optionally, the device further includes a numerical acquisition module, used to input the output of the growth state prediction model to the first training sample set and the first training sample into a preset verification algorithm to obtain training values; and to terminate the model training of the growth state prediction model when the training values ​​exceed a set benchmark value.

[0071] Optionally, the second obtaining module 230 is further configured to input the growth state prediction result into the second input layer to obtain the growth state label and the multimodal features of the facility growth environment; input the growth state label and the multimodal features into the retrieval layer to obtain the retrieval result; and input the retrieval result into the generation layer to obtain the measure text, which is used as the evaluation result.

[0072] Optionally, the device further includes a second training module for acquiring a second training sample set; and training the evaluation model using the second training sample set to obtain the trained evaluation model.

[0073] <Equipment Example 2> Figure 3 This is a schematic diagram of the hardware structure of an electronic device according to another embodiment.

[0074] like Figure 3 As shown, the electronic device 300 includes a processor 310 and a memory 320, the memory 320 being used to store an executable computer program, and the processor 310 being used to execute methods as described in any of the above method embodiments under the control of the computer program.

[0075] Each module of the plant growth prediction device 200 based on the large growth state prediction model described above can be implemented by the processor 310 in this embodiment executing the computer program stored in the memory 320, or it can be implemented by other structures, which are not limited here.

[0076] This invention can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the invention.

[0077] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0078] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0079] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.

[0080] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0081] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0082] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0083] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. It will be known to those skilled in the art that implementation in hardware, implementation in software, and implementation using a combination of software and hardware are equivalent.

[0084] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, and are not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of the invention is defined by the appended claims.

Claims

1. A plant growth prediction method based on a large-scale growth state prediction model, characterized in that, The method includes: The detection device acquires feature detection data of the target plant object and environmental detection data of the growth environment of the facility where the target plant object is located. The feature detection data and the environmental detection data are input into a pre-set large growth state prediction model to obtain the growth state prediction result of the target plant object. The growth state prediction results are input into a preset evaluation model to obtain the evaluation results; Based on the assessment results, the operating parameters of the environmental control device configured for the growth environment of the facility are adjusted.

2. The method according to claim 1, characterized in that, The large-scale growth state prediction model includes a first input layer, a feature fusion layer, and a prediction layer; The step of inputting the feature detection data and the environmental detection data into a pre-set large-scale growth state prediction model to obtain the growth state prediction result of the target plant object includes: The feature detection data and the environmental detection data are input into the first input layer to obtain the instruction text for the target plant object and the multimodal features of the facility's growth environment; The instruction text and the multimodal features are input into the feature fusion layer to obtain the embedding vector; The embedding vector is input into the prediction layer to obtain a binary classification probability, which is used as the prediction result of the growth status of the target plant object.

3. The method according to claim 1, characterized in that, The method further includes: Obtain the first training sample set; The large-scale growth state prediction model is trained using the first training sample set to obtain the trained large-scale growth state prediction model.

4. The method according to claim 3, characterized in that, After training the large-scale growth state prediction model using the first training sample set to obtain the trained large-scale growth state prediction model, the method further includes: The output of the growth state prediction model to the first training sample set and the first training sample are input into a preset verification algorithm to obtain training values. If the training values ​​exceed the set baseline value, the training of the large growth state prediction model is terminated.

5. The method according to claim 1, characterized in that, The evaluation model includes a second input layer, a retrieval layer, and a generation layer; The step of inputting the growth state prediction result into a preset evaluation model to obtain the evaluation result includes: The growth state prediction results are input into the second input layer to obtain the growth state label and the multimodal features of the facility's growth environment; The growth state label and the multimodal features are input into the retrieval layer to obtain the retrieval results; The search results are input into the generation layer to obtain the measure text, which serves as the evaluation result.

6. The method according to claim 1, characterized in that, The method further includes: Obtain the second training sample set; The evaluation model is trained using the second training sample set to obtain the trained evaluation model.

7. The method according to claim 1, characterized in that, The second training sample set includes a first training data set with real labels and a second training data set with pseudo-labels; The second training data set is constructed from the first training data set.

8. A plant growth prediction device based on a large-scale growth state prediction model, characterized in that, The device includes: The acquisition module is used to acquire feature detection data of the target plant object and environmental detection data of the growing environment of the facility where the target plant object is located through the detection device; The first obtaining module is used to input the feature detection data and the environmental detection data into a preset growth state prediction model to obtain the growth state prediction result of the target plant object; The second obtaining module is used to input the growth state prediction results into a preset evaluation model to obtain the evaluation results; The adjustment module is used to adjust the operating parameters of the environmental regulation device configured for the growth environment of the facility based on the evaluation results.

9. An electronic device, characterized in that, The system includes a memory and a processor, the memory being used to store a computer program; the processor being used to execute the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images