Sugarcane germplasm resource evaluation and breeding method for smart agriculture

Abstract

The invention discloses a sugarcane germplasm resource evaluation and breeding method for smart agriculture. The sugarcane germplasm resource evaluation and breeding method comprises the following steps: step 1, collecting materials with wide genetic diversity; establishing an intelligent incubator and a greenhouse, and monitoring environmental parameters by using an Internet of Things sensor; 2, performing data acquisition on the related germplasm resources by using an unmanned aerial vehicle, a robot and a near infrared spectrum technology, and constructing a phenotype database for data management; meanwhile, genotype identification is carried out by combining molecular marker-assisted selection and genome selection technologies, molecular markers of key genes are mined, and the genetic value of related germplasm is evaluated; step 3, constructing a hybrid combination prediction model by using an AI algorithm, optimizing parent matching of the related germplasm and the cultivated sugarcane, monitoring and collecting the filial generation at the same time, and analyzing and screening in combination with smart agriculture; 4, demonstration planting is conducted on the bred excellent sugarcane strain, and application of the excellent strain is improved through government-enterprise-scientific research institution collaborative popularization and in combination with farmer technical training.

Description

Technical Field

[0001] The present invention relates to the technical field of sugarcane germplasm resource evaluation and breeding, and in particular to a sugarcane germplasm resource evaluation and breeding method for smart agriculture. Background Art

[0002] Sugarcane is a tall herb widely cultivated in tropical and subtropical regions. Known for its high sugar content, it is not only the main raw material for the sugar industry, but can also be used to produce biofuels such as ethanol, and has extremely high economic value. Major sugarcane producing areas around the world and in China, such as Guangxi and Yunnan, continue to promote the development of the sugarcane industry by continuously optimizing planting techniques and variety selection. The evaluation and selection of sugarcane germplasm resources are the core driving force behind the progress of this industry.

[0003] However, in the current evaluation and breeding of sugarcane germplasm resources, the utilization of closely related germplasm is still insufficient, resulting in a narrow genetic basis, which restricts the in-depth exploration of key traits such as disease resistance and stress resistance and the further improvement of breeding efficiency. For example, although the thin-stemmed wild species has excellent traits such as stress resistance and tillering, its genetic background is complex (large ploidy variation) and it is difficult to utilize. Therefore, a method for the evaluation and breeding of sugarcane germplasm resources for smart agriculture is proposed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a method for evaluating and breeding sugarcane germplasm resources for smart agriculture.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A method for evaluating and breeding sugarcane germplasm resources for smart agriculture, comprising: Step 1: Extensively collect closely related germplasm resources within the Saccharum complex, including Saccharum officinale, Saccharum variegatum, Saccharum spp., and Saccharum yunnanensis. Through field surveys, international collaboration, and germplasm exchange, collect materials with broad genetic diversity. Secondly, establish intelligent incubators and greenhouses to simulate the native habitat of closely related germplasm resources. Use IoT sensors to monitor environmental parameters in real time, and automatically adjust them through AI algorithms to maintain the genetic stability of the germplasm resources. Simultaneously, use tissue culture seedling technology for asexual reproduction to provide high-quality, genetically pure germplasm materials for subsequent evaluation and breeding. Step 2: Use drones, robots, and near-infrared spectroscopy to collect high-throughput phenotypic data on related germplasm resources, covering agronomic traits, stress resistance, and quality indicators. Build a phenotypic database for real-time data management. Simultaneously, combine marker-assisted selection (MAS) and genomic selection (GS) technologies to conduct genotyping, focusing on identifying molecular markers for key genes related to disease resistance, high sugar content, and stress tolerance. Through the in-depth integration of phenotypic and genotypic data, evaluate the genetic value of related germplasm resources, and provide a scientific basis for parent selection and hybridization breeding. Step 3: Based on the phenotypic and genotypic data from Step 2, an AI algorithm is used to construct a hybrid combination prediction model. Parent selection for closely related germplasms and cultivated sugarcane is optimized based on these phenotypic and genotypic data to improve hybridization efficiency. Dynamic monitoring and data collection are also conducted on hybrid offspring, which are analyzed in real time using the smart agriculture platform. Rigorous screening is then conducted based on pre-set agronomic traits, stress resistance, and quality standards to select new lines with both excellent traits and genetic stability, providing high-quality seed sources for demonstration and promotion in Step 4. Step 4: Carry out demonstration planting of the selected excellent sugarcane varieties on the smart farm, and use smart agricultural technologies such as intelligent water and fertilizer integration and pest and disease monitoring to improve planting management efficiency. At the same time, continuously collect and analyze planting data through the big data platform to dynamically optimize variety performance; finally, through the "government-enterprise-research institution" collaborative promotion model, combined with farmer technical training, improve the application of excellent varieties in the industry, and feedback to step one to continuously enrich the germplasm resource library.

[0006] The above technical solution further includes: Furthermore, the collection of related germplasm resources of the Saccharum complex, through field investigations, international cooperation and germplasm exchange, to collect materials with a wide range of genetic diversity, includes the following steps: Set goals: Identify the range of closely related species within the Saccharum complex, including Echinops, Echinops variegata, Echinops spp., and Echinops yunnanensis, with a focus on selecting germplasm with characteristics such as stress resistance (drought and flood resistance), high quality (high sugar and fiber), and high biomass. Set quantitative indicators, such as the allelic richness of collected materials exceeding the average level in the target region and genetic distance covering the entire range of the complex, to avoid a narrow genetic base for closely related germplasm. Collecting in situ resources: Prioritize areas with rich ecological diversity (such as the tropical / subtropical monsoon regions of Yunnan and Guangxi, China, and the native range of sugarcane in Southeast Asia). Combine historical survey data and literature to mark potential distribution areas (such as riverbanks, hillsides, wetlands, and other different habitats). Organize a team of botanists, agronomists, and local guides (familiar with the terrain and species distribution). Record plant height, leaf shape, stem diameter, and flower spike characteristics to preliminarily screen target germplasm. Collect stem tips, leaves, or seeds (for sexually reproducing species) from healthy plants, collecting at least three plants per collection to ensure genetic representativeness. Use GPS to locate collection points, take photos of the plants (including habitat background), and record soil type, climate data (temperature, precipitation), and other environmental parameters. Introducing high-quality foreign germplasm: Priority should be given to research institutions in leading sugarcane research countries (such as Brazil, India, and Australia) and international germplasm banks (such as CIAT and ILRI). The scope of cooperation should be clearly defined, including germplasm exchange, joint research, and technology sharing (such as disease resistance gene identification technology). The ownership of the germplasm, the scope of use (research / breeding only), and quarantine requirements (to avoid the introduction of foreign pests and diseases) should be clearly defined. The seeds and tissue culture seedlings should be delivered by international express delivery or by experts (which must comply with international plant quarantine standards). Introduced materials must undergo quarantine (for example, tissue culture seedlings must be observed in intelligent incubators for one to two generations to confirm they are virus-free). Germplasm exchange: Select a sugarcane germplasm resource nursery (e.g., the Chinese Academy of Tropical Agricultural Sciences or the Guangxi Sugarcane Research Institute). Evaluate the nursery's preservation capacity (e.g., tissue culture facilities, cryogenic storage capacity) and data integrity (whether phenotypic / genotypic data are available). Submit an application to the nursery, specifying the desired germplasm characteristics (e.g., drought resistance, high sugar content) and quantity (e.g., 10 samples / variety). Acquire resources based on the principle of "equal exchange" (providing specialized germplasm, such as cold-resistant materials) or through scientific research collaboration (joint evaluation). Exchanged germplasm must be entered into a unified database, annotated with information on the source, collection date, and trait descriptions. Supplementary collection for genetic diversity verification: Collected materials are classified and identified (e.g., by flower spike structure, stem node color, etc.), duplicate or incorrectly named germplasm is eliminated, DNA is extracted, genetic diversity is analyzed using SSR (simple sequence repeat) and SNP (single nucleotide polymorphism) markers, genetic distances between materials are calculated, and additional germplasm with distant genetic distances is collected (e.g., if insufficient materials for a certain group are found through cluster analysis); a "collection-verification-supplementation" cycle is established, and collection priorities are adjusted annually based on breeding needs (e.g., the collection of salt- and alkali-tolerant materials is added); Post-collection processing: Short-term storage: The collected living materials (such as stem segments) should be immediately placed in a moisture-retaining bag and transported back to the laboratory within 24 hours for tissue culture seedling induction (using MS medium + plant growth regulators); Long-term storage: After the tissue culture seedlings have been asexually propagated for more than three generations to confirm genetic stability, some are moved to intelligent incubators (simulating the native environment) for long-term storage, and some are transferred to low-temperature storage (-196℃ liquid nitrogen) for backup; Data entry: Enter the morphological data, molecular marker results, environmental parameters, etc. of the collected materials into the germplasm resource database and link them to subsequent evaluation data (such as the phenotypic / genotypic data in step 2).

[0007] Furthermore, the establishment of intelligent incubators and greenhouses to simulate the native environment of closely related germplasms, using IoT sensors to monitor environmental parameters, and automatically adjusting through AI algorithms to maintain the genetic stability of germplasm resources includes the following steps: Investigation of in situ environmental parameters: Restore the temperature, humidity, light, and soil environmental parameters of the native habitat of closely related germplasms to provide a regulatory basis for intelligent breeding. During the "field survey" phase of step one, simultaneously record the environmental parameters of the native habitat of the target germplasm (e.g., the average annual temperature of 20-25°C, relative humidity of 70-85%, and sunshine duration of 10-12 hours / day in the tropical monsoon rain forest of Yunnan). Consult academic literature (e.g., "Studies on the Native Habitat of Closely Related Sugarcane Germplasms in China") or international databases (e.g., the FAO Plant Genetic Resources Database) to supplement parameters that cannot be obtained through field surveys (e.g., soil pH and organic matter content). Classify the collected parameters by species (e.g., Ampelopsis serrata, Ampelopsis serrata), and use statistical methods (e.g., principal component analysis) to screen key environmental factors and establish a "species-environmental parameter" correspondence model (e.g., Ampelopsis serrata requires high humidity and low light, while Ampelopsis serrata requires strong light and large temperature fluctuations). Design and build smart incubators and greenhouses: Incubator: Suitable for small amounts of germplasm (such as tissue culture seedlings), it must be airtight, have adjustable lighting (LED light source), and a temperature and humidity control module (semiconductor heating / cooling); Greenhouse: Suitable for large-scale germplasm conservation, requiring sunshade nets, wet curtain-fan cooling modules, geothermal heating pipes, and CO2 dosing equipment; Deploy IoT sensors: Temperature and humidity sensors (such as DHT22), light sensors (such as BH1750), and soil sensors (such as EC-5 soil moisture sensor) are evenly distributed throughout the incubator and greenhouse. These sensors upload data to edge computing nodes (such as Raspberry Pi) via LoRa or ZigBee protocols, which then transmit the data to a cloud platform (such as Alibaba Cloud IoT) in real time. Develop and train AI algorithms: AI algorithms analyze sensor data and automatically adjust environmental parameters to dynamically match the native habitat model. Using a PID (proportional-integral-derivative) control algorithm, for example, for temperature control, a target temperature (e.g., 25°C) is set. The algorithm then adjusts heating / cooling power based on the difference between the current temperature and the target. Building on PID, an LSTM (long short-term memory) prediction model is introduced. This model combines historical environmental data (e.g., temperature and humidity changes over the past 24 hours) to predict future environmental trends and adjust equipment in advance (e.g., opening sunshade nets in advance when expected sunlight increases the next day). Native habitat survey data (e.g., the annual temperature and humidity curves for a population of Echinops sphagnum mume in Yunnan) is used as a training set to verify the algorithm's accuracy in responding to environmental fluctuations (e.g., temperature control error ≤±0.5°C). Maintaining genetic stability: Collect stem tips (fewer viruses) and leaves (easy to disinfect) from wild germplasm as explants. Adjust the MS culture medium composition according to the species (e.g., for Saccharum officinale, increase the cytokinin concentration to promote cluster bud differentiation). Sterilize the explants in a clean bench (75% alcohol for 30 seconds followed by sodium hypochlorite for 10 minutes). Inoculate the explants into the culture medium and incubate them in the dark at 25°C for one week before switching to light culture (16 hours light / 8 hours dark). Regularly record the plant height, leaf shape, and stem diameter of the tissue culture seedlings and compare them with those in the native habitat to ensure no significant variation. Extract DNA from the tissue culture seedlings and analyze genetic consistency using SSR and SNP markers (e.g., comparing allele frequencies between tissue culture seedlings and wild plants; the difference should be ≤5%). Dynamic Adjustment: The cloud platform displays environmental parameters (temperature, humidity, and light) and equipment status (whether heating / cooling is running) in real time. Abnormal data (such as temperatures exceeding 30°C) triggers alarms and notifies administrators. An AI algorithm analyzes sensor data every 10 minutes and automatically adjusts equipment (such as opening sunshades when light is too strong). It also updates environmental model parameters based on seasonal changes in the native habitat (for example, Yunnan has high humidity during the rainy season and dryness during the dry season), such as increasing sprinkler irrigation frequency in the rainy season and reducing ventilation in the dry season. Data Management: After genetic stability is confirmed through more than three generations of tissue culture, the tissue culture seedlings are moved into an intelligent incubator for long-term storage (the culture medium is renewed every six months), or transferred to a low-temperature storage (-196°C liquid nitrogen) for backup. The environmental parameters during the cultivation process (temperature, humidity, light), tissue culture seedling growth data (plant height, number of tillers), and genetic test results (molecular markers) are entered into the germplasm resource database and associated with the phenotypic and genotypic data of step two to provide complete background information for subsequent evaluation.

[0008] Furthermore, the use of drones, robots, and near-infrared spectroscopy technology to collect high-throughput phenotypic data on related germplasm resources, covering agronomic traits, stress resistance, and quality indicators, includes the following steps: Preliminary preparation: Identify the phenotypic indicators to be collected (agronomic traits, stress resistance, quality indicators) and technical requirements (such as accuracy and throughput). Select a multispectral / hyperspectral drone (such as the DJI P4 Multispectral) equipped with an RGB camera and a 5-band multispectral sensor (visible light - near-infrared), with a flight altitude of 30-50 meters (balancing resolution and coverage). Choose a field phenotyping robot equipped with a lidar, a high-precision camera, and a robotic arm (capable of gripping leaves for near-infrared detection), with a travel speed of 0.5-1 m / s (to ensure data collection stability). Select a portable near-infrared spectrometer (such as the F750) with a wavelength range of 400-2500 nm and a leaf fixture (to ensure leaves are flat and free of creases during testing). Collecting agronomic traits (UAV + robot collaboration): The drone scanned a large area, designing a flight path based on the experimental field layout (covering the entire germplasm area), setting an overlap rate of 70% (to ensure seamless image stitching), and selecting early morning or evening flight times (to avoid interference from strong light). It collected RGB images for plant height measurement (calculated using a 3D reconstruction algorithm) and canopy cover analysis (percentage of green pixels). Multispectral images were also collected to calculate the Normalized Difference Vegetation Index (NDVI) and Leaf Area Index (LAI) to assess plant growth potential. The robot uses drone imagery to locate areas requiring resurvey (such as marginal plants or severely obscured areas). The robot then drives along a pre-set path, using lidar scanning to construct a 3D point cloud model of the plant. This allows for precise calculation of stem diameter and tiller number (main stem and tiller numbers are distinguished by point cloud density). High-definition cameras, combined with AI image recognition (such as the YOLOv5 model), automatically identify leaf count, leaf shape (aspect ratio), and detailed stem node color. Detection of stress resistance indicators (environmental stress + real-time monitoring): Some accessions were drought-treated in the greenhouse (irrigation was stopped for 7-10 days), while a control group (normal irrigation) was maintained. Drone thermal infrared imaging was used to measure leaf temperature (leaf temperature increased under drought stress), and the canopy temperature-air temperature difference (CTD) was calculated. A robotic soil sensor (EC-5 sensor) monitored soil moisture content in real time and correlated it with the degree of plant wilting (leaf curling captured by the robot's camera). The target accessions were inoculated with common sugarcane pathogens (such as Fiji disease virus and rust spores), and isolation areas were set up to prevent cross-infection. Drones identified lesions using characteristic disease bands (such as a decrease in reflectivity near 680nm) and calculated the disease index (percentage of lesion area). Robots also detected pathogen spectral characteristics at the leaf level (such as changes in reflectivity in the 700-800nm ​​band) to identify latent infections at an early stage. Detection quality indicators (near infrared spectroscopy technology): Mature leaves from different accessions (avoiding the veins) were collected, freeze-dried, crushed, and sieved (80 mesh) as standard samples. These samples were scanned using a near-infrared spectrometer (3 scans / sample, with a wavelength interval of 2 nm), and quality indicators such as sugar content (HPLC) and fiber content (Fan's washing method) were simultaneously measured. Partial least squares regression (PLSR) was used to establish a predictive model linking spectral data and quality indicators, with a validation set R² greater than 0.85 to ensure prediction accuracy. A portable near-infrared spectrometer was used to hold live leaves (without destroying the plant), capturing spectra within 3 seconds. The data was then transmitted to a mobile app via Bluetooth, displaying predicted sugar and fiber content in real time. Quality indicators were then correlated with agronomic traits (such as plant height) and stress tolerance (such as drought resistance index) to screen for superior accessions characterized by high yield, high quality, and stress tolerance. Data Integration: Abnormal data (such as blurry drone images and faulty robot sensor values) is eliminated, and median filtering or wavelet transform is used for denoising. The data format is standardized (for example, the resolution of drone RGB images is adjusted to 10 cm / pixel and the robot lidar point cloud density is adjusted to 100 points / cm²). The cleaned data is stored (e.g., in a MySQL database) by "germplasm ID-collection time-indicator type" and associated with the germplasm source information from step 1 (collection location, genetic diversity score). Preliminary data analysis (such as calculating the mean NDVI and sugar content prediction) is performed using edge computing nodes (e.g., Raspberry Pis deployed in the field). Abnormal data (such as a sudden drop in NDVI) triggers an alarm, notifying researchers for on-site verification. Dynamic Updates: Repeatedly collect phenotypic data during the key growth stages of sugarcane (tillering, elongation, and maturity), construct a dynamic growth curve (such as the trend of plant height change over time), and regularly update the near-infrared spectroscopy prediction model based on the newly collected data (retrain once every quarter) to ensure the model's adaptability to different growth stages; upload the desensitized phenotypic data to the National Agricultural Science Data Center for use by other research teams (data use application is required).

[0009] Furthermore, the combination of molecular marker-assisted selection and genomic selection technology to carry out genotype identification and mine molecular markers of key genes for disease resistance, high sugar content, and stress resistance includes the following steps: Sample selection: From the closely related germplasms collected in step 1, select accessions with extreme phenotypes (e.g., extreme drought resistance, high sugar content) as core samples (e.g., 20 accessions for drought resistance and 15 accessions for high sugar content). Build a diversity panel that includes accessions from different species (e.g., Saccharum spp., Saccharum officinale) and geographic origins to ensure genetic background coverage of the entire Saccharum complex. Molecular marker development (MAS basis): Development of known gene markers: Disease resistance genes: Refer to literature (e.g., "Development of Molecular Markers for the Sugarcane Fiji Disease Resistance Gene Bru1") to design PCR-based SSR or InDel markers. Verify primer specificity (e.g., by detecting band uniqueness through agarose gel electrophoresis). High-sugar genes: Utilize the sugarcane transporter gene sequence (e.g., the CDS region of the SugE gene) to design SNP markers (e.g., TaqMan probe method) and detect allelic variations by real-time quantitative PCR (qPCR); New gene marker mining: Genome-wide association study (GWAS): Perform whole-genome resequencing of the core samples (e.g., Illumina NovaSeq platform, 10× coverage). Combined with phenotypic data (e.g., drought resistance index and sugar content in step 2), use PLINK software to screen for SNP markers significantly associated with the target trait (p < 1e-5). Map the associated SNPs to the sugarcane reference genome (e.g., the Saccharum spontaneum genome) and use BLAST to align them with known functional genes (e.g., the WRKY family of stress-resistance-related transcription factors) to identify candidate genes; Identify genotype: For known gene markers, sample DNA is extracted and PCR amplified (annealing temperature 55-60°C). Fragment size is measured using capillary electrophoresis (e.g., ABI 3730xl) to determine the allele type (e.g., "Bru1 disease resistance allele present" or "absent"). For high-sugar gene SNP markers, TaqMan probes (FAM / VIC labeled) are designed, and the fluorescence signal is read using a qPCR instrument (e.g., QuantStudio 5) to determine the SNP locus genotype (e.g., "AA," "AT," "TT"). A sugarcane-specific SNP array (e.g., 50K SNP array) is used to cover evenly distributed SNP loci across the genome, and genotype data are read through hybridization reactions (e.g., Illumina Infinium platform). For samples not arrayed, GBS technology (e.g., ApeKI digestion) is used, and sequencing is performed on the Illumina platform. SNP calling is performed using USTACKS software to generate a genotype matrix. Constructing a genomic selection (GS) model: Select germplasms with complete phenotypic data (agronomic, stress resistance, and quality indicators from step 2) and genotypic data (GS typing results) as the training set (e.g., 100 accessions), and group them by traits (e.g., drought resistance, high sugar content) for modeling; Training the model: Linear model: RR-BLUP (best linear unbiased prediction) model was used, which treated the SNP effect as a random effect and solved the predicted value by a mixed linear equation; Machine learning model: Introduce random forest (RF) or support vector machine, and optimize hyperparameters (such as tree depth and kernel function) through cross-validation (such as 5-fold validation) to improve prediction accuracy (using the correlation coefficient r between phenotype and predicted value as an indicator, with a target r ≥ 0.7); Model validation: Use an independent validation set (e.g., 50 samples not included in training) to test the model's predictive ability and calculate the mean square error (MSE) and r-value between the predicted and measured values ​​to ensure model stability. Key gene mining: MAS marker association analysis statistically analyzes the association between different genotypes (e.g., Bru1 presence / absence) and target traits (e.g., disease resistance) (chi-square test or t-test), and screens for significantly associated markers (p < 0.05). Using SNP weight analysis based on the GS model, we identify gene regions highly associated with the target trait (e.g., SNPs in the top 1% of absolute weights). Combined with GWAS results, we identify candidate genes (e.g., the stress resistance-related gene SnRK2). Overexpressing or knocking out candidate genes (e.g., SnRK2) to observe whether the stress resistance of transgenic plants (e.g., survival rate after drought treatment) is significantly altered. Using CRISPR / Cas9 technology to edit candidate genes (e.g., SugE), we test whether the sugar content of the edited plants decreases to confirm gene function. Data Integration: The genotype data of MAS and GS (including known marker genotypes and GS predicted values) are stored in the format of "germplasm ID-marker name-genotype" (e.g., MongoDB database), and the phenotypic data of step 2 are associated with the germplasm source information of step 1. In hybrid breeding (step 3), parents carrying disease-resistant (Bru1+) and high-sugar (SugE+) alleles are given priority. For hybrid offspring (e.g., F1 generation), their adult phenotypes (e.g., sugar content, drought resistance) are predicted using the GS model, allowing for early screening of excellent individual plants (top 20% of predicted values) to reduce field testing costs.

[0010] Furthermore, the method of constructing a hybrid combination prediction model using an AI algorithm and optimizing parental matching of closely related germplasms and cultivated sugarcane based on phenotypic and genotypic data includes the following steps: Data preprocessing: Integrate the data from Step 1 (conservation of closely related germplasm) and Step 2 (phenotypic / genotypic evaluation) into a structured dataset as input for the AI ​​model. This data includes agronomic traits (plant height, tiller number), stress resistance (drought resistance index, disease resistance rate), and quality indicators (sugar content, fiber content) collected in Step 2; genotypes of molecular markers developed in Step 2 (such as the Bru1 disease resistance marker and the SugE high-sugar marker), and trait values ​​predicted by the genomic selection (GS) model (such as GS drought resistance prediction value and GS sugar prediction value); and phenotypic and genotypic data of existing major cultivated varieties (such as ROC22 and GT42) (collected through the same process to ensure data comparability). KNN interpolation was used for phenotypic data (e.g., missing data from a retest); mean filling was used for genotypic data (e.g., SNP missing rate <5%); Z-score normalization was performed on continuous phenotypic data (e.g., plant height); one-hot encoding was performed on genotypic data (e.g., SNP coded as 0 / 1 / 2); highly redundant features were removed through correlation analysis (Pearson correlation coefficient) (e.g., if the correlation between plant height and stem diameter was >0.8, only one of them was retained); and the top 20 features (e.g., drought resistance index, Bru1 marker, GS sugar content prediction value) were retained through random forest importance ranking. Building an AI model (hybrid combination prediction): Random forest (RF) or gradient boosted tree (GBDT) are used because they are better adaptable to high-dimensional data (genotype SNPs) and small sample sizes (limited number of closely related germplasms). Convolutional neural networks (CNN) are introduced to treat genotype data as a "one-dimensional sequence" (such as a SNP matrix) and extract local features (such as disease resistance-related gene regions) through convolutional layers. Define input and output: Input: phenotypic-genotypic joint characteristics of the parental combination (e.g., drought resistance index of father A + Bru1 marker of mother B + predicted GS sugar value of father A); Output: Expected trait values ​​of hybrid offspring (e.g., sugar content, drought resistance index, disease resistance rate of the F1 generation), which must be defined as continuous values ​​(regression tasks) or categorical values ​​(e.g., "disease resistant / susceptible"); Historical hybridization data (e.g., established hybrid combinations of closely related germplasm and cultivated sugarcane) were used as the training set, recording parental characteristics and offspring traits. Mean squared error (MSE) was used for regression tasks, and cross-entropy loss was used for classification tasks. Model parameters (e.g., n_estimators for RF and learning_rate for GBDT) were adjusted through Bayesian optimization or random search, with MSE and accuracy on the validation set (20% of the data) being used as indicators. Multi-objective optimization: Main goal: maximize comprehensive traits of offspring (e.g., "sugar content ≥ 14% and drought resistance index ≥ 0.8"); Secondary goal: balance other traits (e.g., tiller number ≥ 5, disease resistance ≥ 90%) to avoid over-optimization of a single trait; The NSGA-II (Non-Dominated Sorting Genetic Algorithm) optimization algorithm was selected to generate parent combinations through genetic operations (selection, crossover, and mutation), retaining non-dominated solutions (i.e., no other solution was superior to all other solutions in terms of all objectives). Penalties were added (e.g., if the sugar content of the offspring was less than 14%, the fitness value was reduced) to ensure that the solution met the breeding requirements. Randomly generate 100 parent combinations (closely related germplasm × cultivated sugarcane), each containing the eigenvectors of both the paternal and maternal parents. For each combination, use an AI model to predict offspring traits and calculate fitness (e.g., a combined score of sugar content × drought resistance index × disease resistance rate). A tournament selection method is used to retain combinations with high fitness. A single-point crossover (e.g., swapping a portion of the paternal and maternal genotypes) is performed on the parental eigenvectors. Parental characteristics are randomly altered with a 5% probability (e.g., replacing them with genotypes from other closely related germplasms). The selection process is terminated when fitness has not improved for 10 consecutive generations, and the non-dominated solution set (i.e., the optimal parental combination) is output. Verification optimization: Using K-fold cross-validation (K=5), the data was divided into a training set and a validation set. The average mean square error (MSE) (for regression) and accuracy (for classification) of the model on the validation set were calculated to ensure model stability (e.g., MSE < 0.05). The model's prediction accuracy was tested using new hybrid combinations that were not used in training (e.g., the 2024 hybridization of closely related germplasm × cultivated sugarcane), and the correlation coefficient between the predicted and measured values ​​was calculated (r ≥ 0.7 for acceptance). Based on the test results, model features (e.g., adding soil microbiome data) and algorithm parameters (e.g., increasing the number of NSGA-II iterations) were adjusted to improve prediction accuracy. Hybridization experiments: Based on the optimization results, output the top 20 parent combinations (e.g., "Male parent: Echinops arundinaceae A (drought resistance index 0.9, Bru1+) × Female parent: ROC22 (sugar content 15%, SugE+)") and annotate the expected offspring traits of each combination (e.g., sugar content 14.5%, drought resistance index 0.85, disease resistance 95%). Set up a hybrid isolation zone in a greenhouse or net house to prevent pollen contamination. Artificially emasculate related germplasm (e.g., Echinops arundinaceae A), bag it, and pollinate it with pollen from cultivated sugarcane (ROC22). Mark the hybridization date and combination. After harvesting, the hybrid seeds are dried and disinfected (soaked in sodium hypochlorite), and sown in an intelligent incubator (the facility in step 1) for germination testing (germination rate ≥ 80% is considered acceptable). Dynamic Updates: The phenotypic data of the hybrid F1 generation in step 2 (such as plant height and sugar content) are collected and compared with the model prediction values. The prediction error is calculated (for example, the sugar prediction error is ≤0.5% to be qualified). The newly verified hybrid combination data are included in the training set, and the model is retrained every quarter to improve the prediction ability of new combinations (such as new closely related germplasm × new cultivars); the excellent hybrid offspring (such as the F1 generation with a sugar content of 14.5% and drought resistance) are included in the germplasm resource library in step 1, and the genetic diversity data is updated to form an "evaluation-breeding-conservation" closed loop.

[0011] Furthermore, the dynamic monitoring and data collection of hybrid offspring, combined with real-time analysis on the smart agriculture platform, and screening based on preset agronomic traits, stress resistance and quality standards include the following steps: Smart agriculture platform real-time analysis: Raspberry Pi nodes deployed in the field perform preliminary cleaning (denoising, filling missing values) and real-time analysis (such as calculating NDVI and LAI) of sensor data (temperature, humidity, and light). Based on Alibaba Cloud's ECS server, the Hadoop / Spark big data framework is integrated to store all data (images, spectra, and sensor data) and perform real-time analysis using the Flink stream processing engine. Agronomic traits (plant height, tiller number), stress resistance (CTD, disease index), and quality indicators (sugar content, fiber content) of the same plant are associated with a unique ID to construct an individual-level data archive. For each hybrid offspring, a growth curve model is established (such as the trend of plant height over time). The LSTM algorithm is used to predict adult traits (such as final plant height and sugar content). Thresholds are set (for example, a plant height growth rate of <5 cm / week is considered abnormal), triggering alarms and marking plants for retesting. Standard setting: Agronomic traits: plant height greater than 3m (at maturity), number of tillers greater than 5, stem diameter greater than 2cm (to ensure adaptability to mechanical harvesting); Stress resistance: Drought resistance index (CTD) greater than 0.8 (survival rate after drought treatment greater than 80%), disease resistance rate (rust) greater than 90% (lesion area after inoculation <10%); Quality indicators: sugar content greater than 14% (near infrared predicted value), fiber content less than 12% (Fan's method measured value).

[0012] filter: Initial screening (mid-growth phase): During the tillering phase (90 days after sowing), use drone multispectral data to screen plants with an NDVI ≥ 0.7 (strong growth potential) and no lesions. Eliminate plants with an NDVI < 0.5 or lesion area > 5%. Secondary screening (maturity): During the elongation period (180 days after sowing), combining robot LiDAR data (plant height, stem diameter) with near-infrared spectroscopy data (sugar content), retain plants that simultaneously meet the following criteria: plant height greater than 3 meters, stem diameter greater than 2 cm, and sugar content greater than 14%; Final screening (stress resistance test): Plants that pass the re-screening are subjected to drought treatment (no irrigation for 7 days) and pathogen inoculation. Individuals with CTD < 0.8 or disease resistance rate less than 90% are eliminated, and the top 20% excellent lines are finally retained.

[0013] Furthermore, the application of superior strains can be enhanced through collaborative promotion by government, enterprises and scientific research institutions, combined with technical training for farmers, including the following steps: Clarify the responsibilities of the three parties: Establish a coordinated promotion mechanism of "government guidance, enterprise leadership, and scientific research support" to ensure seamless transition of high-quality strains from the laboratory to the field; Government: Formulate promotion policies (such as seedling subsidies and planting insurance), provide financial support (such as agricultural special funds), build promotion platforms (such as regional sugarcane industry alliances), and coordinate cooperation between enterprises and scientific research institutions; Enterprises: Responsible for seedling production and supply (e.g., through factory-scale tissue culture production to ensure strain purity), building demonstration bases (e.g., a 100-mu core demonstration field) to showcase the yield and stress resistance advantages of superior strains; Research institutions: Provide technical support (such as intelligent water and fertilizer integration solutions and pest and disease monitoring models), conduct farmer training (such as intelligent equipment operation and data interpretation), and solve technical problems in planting; Implementation process: Pilot promotion (first year): Priority will be given to selecting 3-5 pilot counties in major sugarcane-producing areas (such as Guangxi and Yunnan), with one demonstration base (≥50 mu) established in each county. Enterprises will produce seedlings (such as tissue-cultured seedlings or healthy stems) that meet local ecological conditions based on strain characteristics (such as high sugar content and drought resistance) provided by scientific research institutions. Scientific research institutions will equip the demonstration bases with intelligent equipment (such as IoT sensors and drones) and develop planting plans (including integrated water and fertilizer parameters and pest and disease control thresholds). Large-scale promotion (from the second year): Through tissue culture plants or seed-stem propagation fields, the company will increase the annual production capacity of successful pilot varieties to over one million plants. Based on pilot data (e.g., yield and stress resistance), the promotion area will be expanded to surrounding counties and cities, forming a "core demonstration-radiation drive" structure. Build a technical training system for farmers: Training in basic planting techniques includes the characteristics of superior varieties (e.g., high-sugar varieties require water control to increase sugar content), planting density (e.g., row spacing of 1.2m, plant spacing of 0.3m), and fertilization timing (e.g., application of stem-strengthening fertilizer during the elongation period). Training in smart device operation includes the use of IoT sensors (e.g., reading temperature and humidity data), drone-based plant protection (e.g., setting spray parameters), and smart water and fertilizer module control (e.g., adjusting EC values). Training in data interpretation and decision-making includes viewing planting data (e.g., soil moisture, pest and disease warnings) through a mobile app and conducting field management based on thresholds (e.g., initiating irrigation when soil moisture is <60%). Soil temperature, humidity, and light sensors (5m×5m grid) are deployed in promotional plots. Data is transmitted to farmers' mobile phone apps via the LoRa network. Companies offer drone-based plant protection services (e.g., charging 50 yuan per mu) or train farmers to operate the system themselves (requiring drone pilot training and a license). Integrated water and fertilizer equipment (e.g., Venturi spreaders) are installed according to water and fertilizer plans developed by research institutions (e.g., N:P:K = 2:1:1). Irrigation time and fertilizer concentration are remotely controlled via an app. Pest and disease monitoring sensors (e.g., spore traps) are deployed, and data is uploaded to research institution platforms. AI models (e.g., CNN) are used to identify disease types (e.g., rust, Fiji disease), and based on early warning information, biopesticides (e.g., Bacillus thuringiensis) or low-toxic chemical pesticides (e.g., tebuconazole) are recommended, with application times set (e.g., after the morning dew has dried). Effect evaluation feedback: Compare the yield (tons / mu), sugar content (%), and fiber content (%) of superior varieties with those of local dominant varieties (such as ROC22), and calculate the yield increase rate and quality improvement. Compile data on drought resistance (survival rate after drought treatment) and disease resistance (rust incidence) of promoted plots and compare them with data from demonstration bases to verify variety stability. Investigate changes in farmers' income after planting superior varieties (such as sugar factory purchase prices and increased income from yield increases), and calculate input-output ratios (such as the comparison of smart device costs and increased production benefits). Submit issues (such as malfunctions of smart water and fertilizer modules or errors in pest and disease identification) through the farmer app, and research institutions will respond and resolve them within 48 hours. Based on promotion data (such as a certain variety performing poorly in acidic soils), adjust parent selection (such as introducing acid-resistant related germplasm) and update the variety in the next year's promotion. Based on farmer feedback (such as the desire to increase voice prompts in the mobile app), optimize smart devices and software (such as adding dialect voice announcements). Sustainable Development: The tripartite cooperation continues: Government: Allocate special funds annually (e.g., 5 million yuan / year) to support enterprises in expanding seedling propagation and scientific research institutions in upgrading technology; Enterprises: 10% of the promotion income will be returned to scientific research institutions for the selection and breeding of new varieties (such as the development of cold-resistant varieties); Research institutions: Launch at least two new superior strains each year and update technical solutions (such as smart device algorithm optimization); Farmer participation: Farmers are encouraged to establish sugarcane planting cooperatives (e.g. 50 households per cooperative), purchase seedlings and intelligent equipment in a unified manner, reduce individual household costs, and allow farmers with large planting scales (e.g. ≥100 mu) and good technical skills to invest in enterprises with technology or land and share in the promotion benefits.

[0014] The present invention has the following beneficial effects: In the present invention, sugarcane related germplasm with extensive genetic diversity is collected through multiple channels such as field surveys and international cooperation, and the intelligent incubator / greenhouse is combined to simulate the native environment, the Internet of Things + AI regulation is used to maintain genetic stability, and tissue culture seedling technology is used to ensure the quality of asexual reproduction; drones, robots, and near-infrared spectroscopy technology are used to carry out high-throughput phenotypic data collection, and molecular markers and genomic technologies are combined to explore key genes for disease resistance, high sugar content, and stress resistance, and the genetic value of related germplasm is evaluated through deep fusion of phenotype and genotype; and an AI hybrid combination prediction model is constructed based on the above data to optimize the matching of related germplasm and cultivated sugarcane parents, dynamically monitor hybrid offspring, and screen excellent strains according to preset standards; management efficiency is improved through demonstration planting, intelligent water and fertilizer, and pest and disease monitoring technology, and the application of excellent strains is accelerated through collaborative promotion between government, enterprises, and research + farmer training, and data feedback continuously enriches the germplasm resource library, forming a "conservation-evaluation-utilization-feedback" closed loop, which effectively solves the problem of insufficient utilization of related germplasm. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a system block diagram of the sugarcane germplasm resource evaluation and breeding method for smart agriculture proposed by the present invention. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] See also Figure 1 As shown, the present invention is a method for evaluating and breeding sugarcane germplasm resources for smart agriculture, comprising: Step 1: Extensively collect closely related germplasm resources within the Saccharum complex, including Saccharum officinale, Saccharum variegatum, Saccharum spp., and Saccharum yunnanensis. Through field surveys, international collaboration, and germplasm exchange, collect materials with broad genetic diversity. Secondly, establish intelligent incubators and greenhouses to simulate the native habitat of closely related germplasm resources. Use IoT sensors to monitor environmental parameters in real time, and automatically adjust them through AI algorithms to maintain the genetic stability of the germplasm resources. Simultaneously, use tissue culture seedling technology for asexual reproduction to provide high-quality, genetically pure germplasm materials for subsequent evaluation and breeding. Step 2: Use drones, robots, and near-infrared spectroscopy to collect high-throughput phenotypic data on related germplasm resources, covering agronomic traits, stress resistance, and quality indicators. Build a phenotypic database for real-time data management. Simultaneously, combine marker-assisted selection (MAS) and genomic selection (GS) technologies to conduct genotyping, focusing on identifying molecular markers for key genes related to disease resistance, high sugar content, and stress tolerance. Through the in-depth integration of phenotypic and genotypic data, evaluate the genetic value of related germplasm resources, and provide a scientific basis for parent selection and hybridization breeding. Step 3: Based on the phenotypic and genotypic data from Step 2, an AI algorithm is used to construct a hybrid combination prediction model. Parent selection for closely related germplasms and cultivated sugarcane is optimized based on these phenotypic and genotypic data to improve hybridization efficiency. Dynamic monitoring and data collection are also conducted on hybrid offspring, which are analyzed in real time using the smart agriculture platform. Rigorous screening is then conducted based on pre-set agronomic traits, stress resistance, and quality standards to select new lines with both excellent traits and genetic stability, providing high-quality seed sources for demonstration and promotion in Step 4. Step 4: Carry out demonstration planting of the selected excellent sugarcane varieties on the smart farm, and use smart agricultural technologies such as intelligent water and fertilizer integration and pest and disease monitoring to improve planting management efficiency. At the same time, continuously collect and analyze planting data through the big data platform to dynamically optimize variety performance; finally, through the "government-enterprise-research institution" collaborative promotion model, combined with farmer technical training, improve the application of excellent varieties in the industry, and feedback to step one to continuously enrich the germplasm resource library.

[0018] In one embodiment, the collecting of related germplasm resources of the Saccharum complex, through field investigations, international cooperation and germplasm exchange, to collect materials with a wide range of genetic diversity, comprises the following steps: Set goals: Identify the range of closely related species within the Saccharum complex, including Echinops, Echinops variegata, Echinops spp., and Echinops yunnanensis, with a focus on selecting germplasm with characteristics such as stress resistance (drought and flood resistance), high quality (high sugar and fiber), and high biomass. Set quantitative indicators, such as the allelic richness of collected materials exceeding the average level in the target region and genetic distance covering the entire range of the complex, to avoid a narrow genetic base for closely related germplasm. Collecting in situ resources: Prioritize areas with rich ecological diversity (such as the tropical / subtropical monsoon regions of Yunnan and Guangxi, China, and the native range of sugarcane in Southeast Asia). Combine historical survey data and literature to mark potential distribution areas (such as riverbanks, hillsides, wetlands, and other different habitats). Organize a team of botanists, agronomists, and local guides (familiar with the terrain and species distribution). Record plant height, leaf shape, stem diameter, and flower spike characteristics to preliminarily screen target germplasm. Collect stem tips, leaves, or seeds (for sexually reproducing species) from healthy plants, collecting at least three plants per collection to ensure genetic representativeness. Use GPS to locate collection points, take photos of the plants (including habitat background), and record soil type, climate data (temperature, precipitation), and other environmental parameters. Introducing high-quality foreign germplasm: Priority should be given to research institutions in leading sugarcane research countries (such as Brazil, India, and Australia) and international germplasm banks (such as CIAT and ILRI). The scope of cooperation should be clearly defined, including germplasm exchange, joint research, and technology sharing (such as disease resistance gene identification technology). The ownership of the germplasm, the scope of use (research / breeding only), and quarantine requirements (to avoid the introduction of foreign pests and diseases) should be clearly defined. The seeds and tissue culture seedlings should be delivered by international express delivery or by experts (which must comply with international plant quarantine standards). Introduced materials must undergo quarantine (for example, tissue culture seedlings must be observed in intelligent incubators for one to two generations to confirm they are virus-free). Germplasm exchange: Select a sugarcane germplasm resource nursery (e.g., the Chinese Academy of Tropical Agricultural Sciences or the Guangxi Sugarcane Research Institute). Evaluate the nursery's preservation capacity (e.g., tissue culture facilities, cryogenic storage capacity) and data integrity (whether phenotypic / genotypic data are available). Submit an application to the nursery, specifying the desired germplasm characteristics (e.g., drought resistance, high sugar content) and quantity (e.g., 10 samples / variety). Acquire resources based on the principle of "equal exchange" (providing specialized germplasm, such as cold-resistant materials) or through scientific research collaboration (joint evaluation). Exchanged germplasm must be entered into a unified database, annotated with information on the source, collection date, and trait descriptions. Supplementary collection for genetic diversity verification: Collected materials are classified and identified (e.g., by flower spike structure, stem node color, etc.), duplicate or incorrectly named germplasm is eliminated, DNA is extracted, genetic diversity is analyzed using SSR (simple sequence repeat) and SNP (single nucleotide polymorphism) markers, genetic distances between materials are calculated, and additional germplasm with distant genetic distances is collected (e.g., if insufficient materials for a certain group are found through cluster analysis); a "collection-verification-supplementation" cycle is established, and collection priorities are adjusted annually based on breeding needs (e.g., the collection of salt- and alkali-tolerant materials is added); Post-collection processing: Short-term storage: The collected living materials (such as stem segments) should be immediately placed in a moisture-retaining bag and transported back to the laboratory within 24 hours for tissue culture seedling induction (using MS medium + plant growth regulators); Long-term storage: After the tissue culture seedlings have been asexually propagated for more than three generations to confirm genetic stability, some are moved to intelligent incubators (simulating the native environment) for long-term storage, and some are transferred to low-temperature storage (-196℃ liquid nitrogen) for backup; Data entry: Enter the morphological data, molecular marker results, environmental parameters, etc. of the collected materials into the germplasm resource database and link them to subsequent evaluation data (such as the phenotypic / genotypic data in step 2).

[0019] In one embodiment, the establishment of an intelligent incubator and greenhouse to simulate the native environment of closely related germplasm, using IoT sensors to monitor environmental parameters, and automatically adjusting through AI algorithms to maintain the genetic stability of germplasm resources includes the following steps: Investigation of in situ environmental parameters: Restore the temperature, humidity, light, and soil environmental parameters of the native habitat of closely related germplasms to provide a regulatory basis for intelligent breeding. During the "field survey" phase of step one, simultaneously record the environmental parameters of the native habitat of the target germplasm (e.g., the average annual temperature of 20-25°C, relative humidity of 70-85%, and sunshine duration of 10-12 hours / day in the tropical monsoon rain forest of Yunnan). Consult academic literature (e.g., "Studies on the Native Habitat of Closely Related Sugarcane Germplasms in China") or international databases (e.g., the FAO Plant Genetic Resources Database) to supplement parameters that cannot be obtained through field surveys (e.g., soil pH and organic matter content). Classify the collected parameters by species (e.g., Ampelopsis serrata, Ampelopsis serrata), and use statistical methods (e.g., principal component analysis) to screen key environmental factors and establish a "species-environmental parameter" correspondence model (e.g., Ampelopsis serrata requires high humidity and low light, while Ampelopsis serrata requires strong light and large temperature fluctuations). Design and build smart incubators and greenhouses: Incubator: Suitable for small amounts of germplasm (such as tissue culture seedlings), it must be airtight, have adjustable lighting (LED light source), and a temperature and humidity control module (semiconductor heating / cooling); Greenhouse: Suitable for large-scale germplasm conservation, it needs to be equipped with sunshade nets, wet curtain-fan cooling system, geothermal heating pipes, and CO2 injection equipment; Deploy IoT sensors: Temperature and humidity sensors (such as DHT22), light sensors (such as BH1750), and soil sensors (such as EC-5 soil moisture sensor) are evenly distributed throughout the incubator and greenhouse. These sensors upload data to edge computing nodes (such as Raspberry Pi) via LoRa or ZigBee protocols, which then transmit the data to a cloud platform (such as Alibaba Cloud IoT) in real time. Develop and train AI algorithms: AI algorithms analyze sensor data and automatically adjust environmental parameters to dynamically match the native habitat model. Using a PID (proportional-integral-derivative) control algorithm, for example, for temperature control, a target temperature (e.g., 25°C) is set. The algorithm then adjusts heating / cooling power based on the difference between the current temperature and the target. Building on PID, an LSTM (long short-term memory) prediction model is introduced. This model combines historical environmental data (e.g., temperature and humidity changes over the past 24 hours) to predict future environmental trends and adjust equipment in advance (e.g., opening sunshade nets in advance when expected sunlight increases the next day). Native habitat survey data (e.g., the annual temperature and humidity curves for a population of Echinops sphagnum mume in Yunnan) is used as a training set to verify the algorithm's accuracy in responding to environmental fluctuations (e.g., temperature control error ≤±0.5°C). Maintaining genetic stability: Collect stem tips (fewer viruses) and leaves (easy to disinfect) from wild germplasm as explants. Adjust the MS culture medium composition according to the species (e.g., for Saccharum officinale, increase the cytokinin concentration to promote cluster bud differentiation). Sterilize the explants in a clean bench (75% alcohol for 30 seconds followed by sodium hypochlorite for 10 minutes). Inoculate the explants into the culture medium and incubate them in the dark at 25°C for one week before switching to light culture (16 hours light / 8 hours dark). Regularly record the plant height, leaf shape, and stem diameter of the tissue culture seedlings and compare them with those in the native habitat to ensure no significant variation. Extract DNA from the tissue culture seedlings and analyze genetic consistency using SSR and SNP markers (e.g., comparing allele frequencies between tissue culture seedlings and wild plants; the difference should be ≤5%). Dynamic Adjustment: The cloud platform displays environmental parameters (temperature, humidity, and light) and equipment status (whether heating / cooling is running) in real time. Abnormal data (such as temperatures exceeding 30°C) triggers alarms and notifies administrators. An AI algorithm analyzes sensor data every 10 minutes and automatically adjusts equipment (such as opening sunshades when light is too strong). It also updates environmental model parameters based on seasonal changes in the native habitat (for example, Yunnan has high humidity during the rainy season and dryness during the dry season), such as increasing sprinkler irrigation frequency in the rainy season and reducing ventilation in the dry season. Data Management: After genetic stability is confirmed through more than three generations of tissue culture, the tissue culture seedlings are moved into an intelligent incubator for long-term storage (the culture medium is renewed every six months), or transferred to a low-temperature storage (-196°C liquid nitrogen) for backup. The environmental parameters during the cultivation process (temperature, humidity, light), tissue culture seedling growth data (plant height, number of tillers), and genetic test results (molecular markers) are entered into the germplasm resource database and associated with the phenotypic and genotypic data of step two to provide complete background information for subsequent evaluation.

[0020] In one embodiment, the high-throughput phenotypic data collection of closely related germplasm resources using drones, robots, and near-infrared spectroscopy, covering agronomic traits, stress resistance, and quality indicators, includes the following steps: Preliminary preparation: Identify the phenotypic indicators to be collected (agronomic traits, stress resistance, quality indicators) and technical requirements (such as accuracy and throughput). Select a multispectral / hyperspectral drone (such as the DJI P4 Multispectral) equipped with an RGB camera and a 5-band multispectral sensor (visible light - near-infrared), with a flight altitude of 30-50 meters (balancing resolution and coverage). Choose a field phenotyping robot equipped with a lidar, a high-precision camera, and a robotic arm (capable of gripping leaves for near-infrared detection), with a travel speed of 0.5-1 m / s (to ensure data collection stability). Select a portable near-infrared spectrometer (such as the F750) with a wavelength range of 400-2500 nm and a leaf fixture (to ensure leaves are flat and free of creases during testing). Collecting agronomic traits (UAV + robot collaboration): The drone scanned a large area, designing a flight path based on the experimental field layout (covering the entire germplasm area), setting an overlap rate of 70% (to ensure seamless image stitching), and selecting early morning or evening flight times (to avoid interference from strong light). It collected RGB images for plant height measurement (calculated using a 3D reconstruction algorithm) and canopy cover analysis (percentage of green pixels). Multispectral images were also collected to calculate the Normalized Difference Vegetation Index (NDVI) and Leaf Area Index (LAI) to assess plant growth potential. The robot uses drone imagery to locate areas requiring resurvey (such as marginal plants or severely obscured areas). The robot then drives along a pre-set path, using lidar scanning to construct a 3D point cloud model of the plant. This allows for precise calculation of stem diameter and tiller number (main stem and tiller numbers are distinguished by point cloud density). High-definition cameras, combined with AI image recognition (such as the YOLOv5 model), automatically identify leaf count, leaf shape (aspect ratio), and detailed stem node color. Detection of stress resistance indicators (environmental stress + real-time monitoring): Some accessions were drought-treated in the greenhouse (irrigation was stopped for 7-10 days), while a control group (normal irrigation) was maintained. Drone thermal infrared imaging was used to measure leaf temperature (leaf temperature increased under drought stress), and the canopy temperature-air temperature difference (CTD) was calculated. A robotic soil sensor (EC-5 sensor) monitored soil moisture content in real time and correlated it with the degree of plant wilting (leaf curling captured by the robot's camera). The target accessions were inoculated with common sugarcane pathogens (such as Fiji disease virus and rust spores), and isolation areas were set up to prevent cross-infection. Drones identified lesions using characteristic disease bands (such as a decrease in reflectivity near 680nm) and calculated the disease index (percentage of lesion area). Robots also detected pathogen spectral characteristics at the leaf level (such as changes in reflectivity in the 700-800nm ​​band) to identify latent infections at an early stage. Detection quality indicators (near infrared spectroscopy technology): Mature leaves from different accessions (avoiding the veins) were collected, freeze-dried, crushed, and sieved (80 mesh) as standard samples. These samples were scanned using a near-infrared spectrometer (3 scans / sample, with a wavelength interval of 2 nm), and quality indicators such as sugar content (HPLC) and fiber content (Fan's washing method) were simultaneously measured. Partial least squares regression (PLSR) was used to establish a predictive model linking spectral data and quality indicators, with a validation set R² greater than 0.85 to ensure prediction accuracy. A portable near-infrared spectrometer was used to hold live leaves (without destroying the plant), capturing spectra within 3 seconds. The data was then transmitted to a mobile app via Bluetooth, displaying predicted sugar and fiber content in real time. Quality indicators were then correlated with agronomic traits (such as plant height) and stress tolerance (such as drought resistance index) to screen for superior accessions characterized by high yield, high quality, and stress tolerance. Data Integration: Abnormal data (such as blurry drone images and faulty robot sensor values) is eliminated, and median filtering or wavelet transform is used for denoising. The data format is standardized (for example, the resolution of drone RGB images is adjusted to 10 cm / pixel and the robot lidar point cloud density is adjusted to 100 points / cm²). The cleaned data is stored (e.g., in a MySQL database) by "germplasm ID-collection time-indicator type" and associated with the germplasm source information from step 1 (collection location, genetic diversity score). Preliminary data analysis (such as calculating the mean NDVI and sugar content prediction) is performed using edge computing nodes (e.g., Raspberry Pis deployed in the field). Abnormal data (such as a sudden drop in NDVI) triggers an alarm, notifying researchers for on-site verification. Dynamic Updates: Repeatedly collect phenotypic data during the key growth stages of sugarcane (tillering, elongation, and maturity), construct a dynamic growth curve (such as the trend of plant height change over time), and regularly update the near-infrared spectroscopy prediction model based on the newly collected data (retrain once every quarter) to ensure the model's adaptability to different growth stages; upload the desensitized phenotypic data to the National Agricultural Science Data Center for use by other research teams (data use application is required).

[0021] In one embodiment, the method of combining molecular marker-assisted selection with genomic selection technology to carry out genotyping identification and mine molecular markers of key genes for disease resistance, high sugar content, and stress resistance includes the following steps: Sample selection: From the closely related germplasms collected in step 1, select accessions with extreme phenotypes (e.g., extreme drought resistance, high sugar content) as core samples (e.g., 20 accessions for drought resistance and 15 accessions for high sugar content). Build a diversity panel that includes accessions from different species (e.g., Saccharum spp., Saccharum officinale) and geographic origins to ensure genetic background coverage of the entire Saccharum complex. Molecular marker development (MAS basis): Development of known gene markers: Disease resistance genes: Refer to literature (e.g., "Development of Molecular Markers for the Sugarcane Fiji Disease Resistance Gene Bru1") to design PCR-based SSR or InDel markers. Verify primer specificity (e.g., by detecting band uniqueness through agarose gel electrophoresis). High-sugar genes: Utilize the sugarcane transporter gene sequence (e.g., the CDS region of the SugE gene) to design SNP markers (e.g., TaqMan probe method) and detect allelic variations by real-time quantitative PCR (qPCR); New gene marker mining: Genome-wide association study (GWAS): Perform whole-genome resequencing of the core samples (e.g., Illumina NovaSeq platform, 10× coverage). Combined with phenotypic data (e.g., drought resistance index and sugar content in step 2), use PLINK software to screen for SNP markers significantly associated with the target trait (p < 1e-5). Map the associated SNPs to the sugarcane reference genome (e.g., the Saccharum spontaneum genome) and use BLAST to align them with known functional genes (e.g., the WRKY family of stress-resistance-related transcription factors) to identify candidate genes; Identify genotype: For known gene markers, sample DNA is extracted and PCR amplified (annealing temperature 55-60°C). Fragment size is measured using capillary electrophoresis (e.g., ABI 3730xl) to determine the allele type (e.g., "Bru1 disease resistance allele present" or "absent"). For high-sugar gene SNP markers, TaqMan probes (FAM / VIC labeled) are designed, and the fluorescence signal is read using a qPCR instrument (e.g., QuantStudio 5) to determine the SNP locus genotype (e.g., "AA," "AT," "TT"). A sugarcane-specific SNP array (e.g., 50K SNP array) is used to cover evenly distributed SNP loci across the genome, and genotype data are read through hybridization reactions (e.g., Illumina Infinium platform). For samples not arrayed, GBS technology (e.g., ApeKI digestion) is used, and sequencing is performed on the Illumina platform. SNP calling is performed using USTACKS software to generate a genotype matrix. Constructing a genomic selection (GS) model: Select germplasms with complete phenotypic data (agronomic, stress resistance, and quality indicators from step 2) and genotypic data (GS typing results) as the training set (e.g., 100 accessions), and group them by traits (e.g., drought resistance, high sugar content) for modeling; Training the model: Linear model: RR-BLUP (best linear unbiased prediction) model was used, which treated the SNP effect as a random effect and solved the predicted value by a mixed linear equation; Machine learning model: Introduce random forest (RF) or support vector machine, and optimize hyperparameters (such as tree depth and kernel function) through cross-validation (such as 5-fold validation) to improve prediction accuracy (using the correlation coefficient r between phenotype and predicted value as an indicator, with a target r ≥ 0.7); Model validation: Use an independent validation set (e.g., 50 samples not included in training) to test the model's predictive ability and calculate the mean square error (MSE) and r-value between the predicted and measured values ​​to ensure model stability. Key gene mining: MAS marker association analysis statistically analyzes the association between different genotypes (e.g., Bru1 presence / absence) and target traits (e.g., disease resistance) (chi-square test or t-test), and screens for significantly associated markers (p < 0.05). Using SNP weight analysis based on the GS model, we identify gene regions highly associated with the target trait (e.g., SNPs in the top 1% of absolute weights). Combined with GWAS results, we identify candidate genes (e.g., the stress resistance-related gene SnRK2). Overexpressing or knocking out candidate genes (e.g., SnRK2) to observe whether the stress resistance of transgenic plants (e.g., survival rate after drought treatment) is significantly altered. Using CRISPR / Cas9 technology to edit candidate genes (e.g., SugE), we test whether the sugar content of the edited plants decreases to confirm gene function. Data Integration: The genotype data of MAS and GS (including known marker genotypes and GS predicted values) are stored in the format of "germplasm ID-marker name-genotype" (e.g., MongoDB database), and the phenotypic data of step 2 are associated with the germplasm source information of step 1. In hybrid breeding (step 3), parents carrying disease-resistant (Bru1+) and high-sugar (SugE+) alleles are given priority. For hybrid offspring (e.g., F1 generation), their adult phenotypes (e.g., sugar content, drought resistance) are predicted using the GS model, allowing for early screening of excellent individual plants (top 20% of predicted values) to reduce field testing costs.

[0022] In one embodiment, the method of constructing a hybrid combination prediction model using an AI algorithm to optimize parental matching of closely related germplasms and cultivated sugarcane based on phenotypic and genotypic data comprises the following steps: Data preprocessing: Integrate the data from Step 1 (conservation of closely related germplasm) and Step 2 (phenotypic / genotypic evaluation) into a structured dataset as input for the AI ​​model. This data includes agronomic traits (plant height, tiller number), stress resistance (drought resistance index, disease resistance rate), and quality indicators (sugar content, fiber content) collected in Step 2; genotypes of molecular markers developed in Step 2 (such as the Bru1 disease resistance marker and the SugE high-sugar marker), and trait values ​​predicted by the genomic selection (GS) model (such as GS drought resistance prediction value and GS sugar prediction value); and phenotypic and genotypic data of existing major cultivated varieties (such as ROC22 and GT42) (collected through the same process to ensure data comparability). KNN interpolation was used for phenotypic data (e.g., missing data from a retest); mean filling was used for genotypic data (e.g., SNP missing rate <5%); Z-score normalization was performed on continuous phenotypic data (e.g., plant height); one-hot encoding was performed on genotypic data (e.g., SNP coded as 0 / 1 / 2); highly redundant features were removed through correlation analysis (Pearson correlation coefficient) (e.g., if the correlation between plant height and stem diameter was >0.8, only one of them was retained); and the top 20 features (e.g., drought resistance index, Bru1 marker, GS sugar content prediction value) were retained through random forest importance ranking. Building an AI model (hybrid combination prediction): Random forest (RF) or gradient boosted tree (GBDT) are used because they are better adaptable to high-dimensional data (genotype SNPs) and small sample sizes (limited number of closely related germplasms). Convolutional neural networks (CNN) are introduced to treat genotype data as a "one-dimensional sequence" (such as a SNP matrix) and extract local features (such as disease resistance-related gene regions) through convolutional layers. Define input and output: Input: phenotypic-genotypic joint characteristics of the parental combination (e.g., drought resistance index of father A + Bru1 marker of mother B + predicted GS sugar value of father A); Output: Expected trait values ​​of hybrid offspring (e.g., sugar content, drought resistance index, disease resistance rate of the F1 generation), which must be defined as continuous values ​​(regression tasks) or categorical values ​​(e.g., "disease resistant / susceptible"); Historical hybridization data (e.g., established hybrid combinations of closely related germplasm and cultivated sugarcane) were used as the training set, recording parental characteristics and offspring traits. Mean squared error (MSE) was used for regression tasks, and cross-entropy loss was used for classification tasks. Model parameters (e.g., n_estimators for RF and learning_rate for GBDT) were adjusted through Bayesian optimization or random search, with MSE and accuracy on the validation set (20% of the data) being used as indicators. Multi-objective optimization: Main goal: maximize comprehensive traits of offspring (e.g., "sugar content ≥ 14% and drought resistance index ≥ 0.8"); Secondary goal: balance other traits (e.g., tiller number ≥ 5, disease resistance ≥ 90%) to avoid over-optimization of a single trait; The NSGA-II (Non-Dominated Sorting Genetic Algorithm) optimization algorithm was selected to generate parent combinations through genetic operations (selection, crossover, and mutation), retaining non-dominated solutions (i.e., no other solution was superior to all other solutions in terms of all objectives). Penalties were added (e.g., if the sugar content of the offspring was less than 14%, the fitness value was reduced) to ensure that the solution met the breeding requirements. Randomly generate 100 parent combinations (closely related germplasm × cultivated sugarcane), each containing the eigenvectors of both the paternal and maternal parents. For each combination, use an AI model to predict offspring traits and calculate fitness (e.g., a combined score of sugar content × drought resistance index × disease resistance rate). A tournament selection method is used to retain combinations with high fitness. A single-point crossover (e.g., swapping a portion of the paternal and maternal genotypes) is performed on the parental eigenvectors. Parental characteristics are randomly altered with a 5% probability (e.g., replacing them with genotypes from other closely related germplasms). The selection process is terminated when fitness has not improved for 10 consecutive generations, and the non-dominated solution set (i.e., the optimal parental combination) is output. Verification optimization: Using K-fold cross-validation (K=5), the data was divided into a training set and a validation set. The average mean square error (MSE) (for regression) and accuracy (for classification) of the model on the validation set were calculated to ensure model stability (e.g., MSE < 0.05). The model's prediction accuracy was tested using new hybrid combinations that were not used in training (e.g., the 2024 hybridization of closely related germplasm × cultivated sugarcane), and the correlation coefficient between the predicted and measured values ​​was calculated (r ≥ 0.7 for acceptance). Based on the test results, model features (e.g., adding soil microbiome data) and algorithm parameters (e.g., increasing the number of NSGA-II iterations) were adjusted to improve prediction accuracy. Hybridization experiments: Based on the optimization results, output the top 20 parent combinations (e.g., "Male parent: Echinops arundinaceae A (drought resistance index 0.9, Bru1+) × Female parent: ROC22 (sugar content 15%, SugE+)") and annotate the expected offspring traits of each combination (e.g., sugar content 14.5%, drought resistance index 0.85, disease resistance 95%). Set up a hybrid isolation zone in a greenhouse or net house to prevent pollen contamination. Artificially emasculate related germplasm (e.g., Echinops arundinaceae A), bag it, and pollinate it with pollen from cultivated sugarcane (ROC22). Mark the hybridization date and combination. After harvesting, the hybrid seeds are dried and disinfected (soaked in sodium hypochlorite), and sown in an intelligent incubator (the facility in step 1) for germination testing (germination rate ≥ 80% is considered acceptable). Dynamic Updates: The phenotypic data of the hybrid F1 generation in step 2 (such as plant height and sugar content) are collected and compared with the model prediction values. The prediction error is calculated (for example, the sugar prediction error is ≤0.5% to be qualified). The newly verified hybrid combination data are included in the training set, and the model is retrained every quarter to improve the prediction ability of new combinations (such as new closely related germplasm × new cultivars); the excellent hybrid offspring (such as the F1 generation with a sugar content of 14.5% and drought resistance) are included in the germplasm resource library in step 1, and the genetic diversity data is updated to form an "evaluation-breeding-conservation" closed loop.

[0023] In one embodiment, the dynamic monitoring and data collection of hybrid offspring, combined with real-time analysis on a smart agriculture platform, and screening based on preset agronomic traits, stress resistance, and quality standards, include the following steps: Smart agriculture platform real-time analysis: Raspberry Pi nodes deployed in the field perform preliminary cleaning (denoising, filling missing values) and real-time analysis (such as calculating NDVI and LAI) of sensor data (temperature, humidity, and light). Based on Alibaba Cloud's ECS server, the Hadoop / Spark big data framework is integrated to store all data (images, spectra, and sensor data) and perform real-time analysis using the Flink stream processing engine. Agronomic traits (plant height, tiller number), stress resistance (CTD, disease index), and quality indicators (sugar content, fiber content) of the same plant are associated with a unique ID to construct an individual-level data archive. For each hybrid offspring, a growth curve model is established (such as the trend of plant height over time). The LSTM algorithm is used to predict adult traits (such as final plant height and sugar content). Thresholds are set (for example, a plant height growth rate of <5 cm / week is considered abnormal), triggering alarms and marking plants for retesting. Standard setting: Agronomic traits: plant height greater than 3m (at maturity), number of tillers greater than 5, stem diameter greater than 2cm (to ensure adaptability to mechanical harvesting); Stress resistance: Drought resistance index (CTD) greater than 0.8 (survival rate after drought treatment greater than 80%), disease resistance rate (rust) greater than 90% (lesion area after inoculation <10%); Quality indicators: sugar content greater than 14% (near infrared predicted value), fiber content less than 12% (Fan's method measured value).

[0024] filter: Initial screening (mid-growth phase): During the tillering phase (90 days after sowing), use drone multispectral data to screen plants with an NDVI ≥ 0.7 (strong growth potential) and no lesions. Eliminate plants with an NDVI < 0.5 or lesion area > 5%. Secondary screening (maturity): During the elongation period (180 days after sowing), combining robot LiDAR data (plant height, stem diameter) with near-infrared spectroscopy data (sugar content), retain plants that simultaneously meet the following criteria: plant height greater than 3 meters, stem diameter greater than 2 cm, and sugar content greater than 14%; Final screening (stress resistance test): Plants that pass the re-screening are subjected to drought treatment (no irrigation for 7 days) and pathogen inoculation. Individuals with CTD < 0.8 or disease resistance rate less than 90% are eliminated, and the top 20% excellent lines are finally retained.

[0025] In one embodiment, the application of improved strains is enhanced through collaborative promotion by government, enterprises, and scientific research institutions, combined with technical training for farmers, including the following steps: Clarify the responsibilities of the three parties: Establish a coordinated promotion mechanism of "government guidance, enterprise leadership, and scientific research support" to ensure seamless transition of high-quality strains from the laboratory to the field; Government: Formulate promotion policies (such as seedling subsidies and planting insurance), provide financial support (such as agricultural special funds), build promotion platforms (such as regional sugarcane industry alliances), and coordinate cooperation between enterprises and scientific research institutions; Enterprises: Responsible for seedling production and supply (e.g., through factory-scale tissue culture production to ensure strain purity), building demonstration bases (e.g., a 100-mu core demonstration field) to showcase the yield and stress resistance advantages of superior strains; Research institutions: Provide technical support (such as intelligent water and fertilizer integration solutions and pest and disease monitoring models), conduct farmer training (such as intelligent equipment operation and data interpretation), and solve technical problems in planting; Implementation process: Pilot promotion (first year): Priority will be given to selecting 3-5 pilot counties in major sugarcane-producing areas (such as Guangxi and Yunnan), with one demonstration base (≥50 mu) established in each county. Enterprises will produce seedlings (such as tissue-cultured seedlings or healthy stems) that meet local ecological conditions based on strain characteristics (such as high sugar content and drought resistance) provided by scientific research institutions. Scientific research institutions will equip the demonstration bases with intelligent equipment (such as IoT sensors and drones) and develop planting plans (including integrated water and fertilizer parameters and pest and disease control thresholds). Large-scale promotion (from the second year): Through tissue culture plants or seed-stem propagation fields, the company will increase the annual production capacity of successful pilot varieties to over one million plants. Based on pilot data (e.g., yield and stress resistance), the promotion area will be expanded to surrounding counties and cities, forming a "core demonstration-radiation drive" structure. Build a technical training system for farmers: Training in basic planting techniques includes the characteristics of superior varieties (e.g., high-sugar varieties require water control to increase sugar content), planting density (e.g., row spacing of 1.2m, plant spacing of 0.3m), and fertilization timing (e.g., applying stem-strengthening fertilizer during the elongation period). Training in smart device operation includes the use of IoT sensors (e.g., reading temperature and humidity data), drone-based plant protection (e.g., setting spray parameters), and smart water and fertilizer system control (e.g., adjusting EC values). Training in data interpretation and decision-making includes viewing planting data (e.g., soil moisture, pest and disease warnings) through a mobile app and conducting field management based on thresholds (e.g., initiating irrigation when soil moisture is <60%). Soil temperature, humidity, and light sensors (5m×5m grid) are deployed in promotional plots. Data is transmitted to farmers' mobile phone apps via the LoRa network. Companies offer drone-based plant protection services (e.g., charging 50 yuan per mu) or train farmers to operate the system themselves (requiring drone pilot training and a license). Integrated water and fertilizer equipment (e.g., Venturi spreaders) are installed according to water and fertilizer plans developed by research institutions (e.g., N:P:K = 2:1:1). Irrigation time and fertilizer concentration are remotely controlled via an app. Pest and disease monitoring sensors (e.g., spore traps) are deployed, and data is uploaded to research institution platforms. AI models (e.g., CNN) are used to identify disease types (e.g., rust, Fiji disease), and based on early warning information, biopesticides (e.g., Bacillus thuringiensis) or low-toxic chemical pesticides (e.g., tebuconazole) are recommended, with application times set (e.g., after the morning dew has dried). Effect evaluation feedback: Compare the yield (tons / mu), sugar content (%), and fiber content (%) of superior varieties with those of local dominant varieties (e.g., ROC22), and calculate the yield increase rate and quality improvement. Compile data on drought resistance (survival rate after drought treatment) and disease resistance (rust incidence) of promoted plots and compare them with data from demonstration bases to verify variety stability. Investigate changes in farmers' income after planting superior varieties (e.g., sugar factory purchase price, increased income from yield increase), and calculate input-output ratios (e.g., comparison of smart device costs with increased yield benefits). Submit issues (e.g., smart water and fertilizer system failures, pest and disease identification errors) through the farmer app, and research institutions will respond and resolve them within 48 hours. Based on promotion data (e.g., a certain variety performs poorly in acidic soils), adjust parent selection (e.g., introduce acid-tolerant related germplasm), and update varieties for the next year's promotion. Based on farmer feedback (e.g., requests for voice prompts on mobile apps), optimize smart devices and software (e.g., add dialect voice announcements). Sustainable Development: The tripartite cooperation continues: Government: Allocate special funds annually (e.g., 5 million yuan / year) to support enterprises in expanding seedling propagation and scientific research institutions in upgrading technology; Enterprises: 10% of the promotion income will be returned to scientific research institutions for the selection and breeding of new varieties (such as the development of cold-resistant varieties); Research institutions: Launch at least two new superior strains each year and update technical solutions (such as smart device algorithm optimization); Farmer participation: Farmers are encouraged to establish sugarcane planting cooperatives (e.g. 50 households per cooperative), purchase seedlings and intelligent equipment in a unified manner, reduce individual household costs, and allow farmers with large planting scales (e.g. ≥100 mu) and good technical skills to invest in enterprises with technology or land and share in the promotion benefits.

[0026] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating and breeding sugarcane germplasm resources for smart agriculture, characterized in that: include: Step 1: Collect closely related germplasm resources of the Saccharum complex. Collect materials with broad genetic diversity through field surveys, international collaboration, and germplasm exchange. Establish intelligent incubators and greenhouses to simulate the native habitat of closely related germplasm. Use IoT sensors to monitor environmental parameters and automatically adjust them through AI algorithms to maintain the genetic stability of the germplasm resources. Simultaneously, use tissue culture seedling technology for asexual propagation to provide high-quality, genetically pure germplasm materials for subsequent evaluation and breeding. Step 2: Use drones, robots, and near-infrared spectroscopy to collect high-throughput phenotypic data on related germplasm resources, covering agronomic traits, stress resistance, and quality indicators, and build a phenotypic database for real-time data management. Simultaneously, combine molecular marker-assisted selection and genomic selection technologies to conduct genotyping, discover molecular markers for key genes for disease resistance, high sugar content, and stress tolerance, and evaluate the genetic value of related germplasm through the deep integration of phenotypic and genotypic data. Step 3: Based on the phenotypic and genotypic data from Step 2, an AI algorithm is used to construct a hybrid combination prediction model. Parent selection for closely related germplasms and cultivated sugarcane is optimized based on these phenotypic and genotypic data to improve hybridization efficiency. Dynamic monitoring and data collection are also conducted on hybrid offspring. Real-time analysis is conducted using the smart agriculture platform. Screening is conducted based on pre-set agronomic traits, stress resistance, and quality standards to select new lines with both excellent traits and genetic stability, providing high-quality seed sources for demonstration and promotion in Step 4. Step 4: Carry out demonstration planting of the selected excellent sugarcane varieties, use intelligent water and fertilizer integration and pest and disease monitoring smart agricultural technologies to improve planting management efficiency, collect and analyze planting data through big data platforms, and dynamically optimize variety performance; through collaborative promotion by government, enterprises, and scientific research institutions, combined with technical training for farmers, improve the application of excellent varieties, and feedback to step one to continuously enrich the germplasm resource library.

2. The method for evaluating and breeding sugarcane germplasm resources for smart agriculture according to claim 1, characterized in that: The method of collecting related germplasm resources of the Saccharum complex, collecting materials with wide genetic diversity through field surveys, international cooperation and germplasm exchange, includes the following steps: Set goals: Identify the range of closely related species within the Saccharum complex, including Echinops, Echinops variegata, Echinops spp., and Echinops yunnanensis, with a focus on selecting germplasm with stress resistance, high quality, and high biomass. Set quantitative targets to ensure that the allelic richness of collected materials is greater than the average level in the target region and that the genetic distance covers the entire range of the complex, to avoid a narrow genetic base for closely related germplasm. Collecting in situ resources: Prioritize areas with rich ecological diversity, combine historical survey data and literature to mark potential distribution areas; organize botanists, agronomists, and local guides to record plant height, leaf shape, stem diameter, and flower spike characteristics to preliminarily screen target germplasm; collect stem tips, leaves, and seeds from healthy plants, and collect at least three plants per collection to improve genetic representativeness; use GPS to locate collection points, take photos of plants, and record soil type, climate data, and environmental parameters; Introducing high-quality foreign germplasm: Priority should be given to scientific research institutions and international germplasm banks in leading sugarcane research countries. The scope of cooperation for germplasm exchange, joint investigations, and technology sharing should be clearly defined. The ownership, scope of use, and quarantine requirements of the germplasm should be clearly defined. The germplasm should be delivered by international express or by experts. Introduced materials must undergo quarantine. Germplasm exchange: Select national and provincial sugarcane germplasm resource nurseries and assess their preservation capacity and data integrity. Submit an application to the nursery, specifying the characteristics and quantity of the germplasm required. Exchange the germplasm on an equal basis, or acquire resources through scientific research collaboration. The exchanged germplasm must be entered into a unified database, annotated with information on the source, collection time, and trait description. Collection and processing: The collected materials are classified and identified, duplicate and incorrectly named germplasm are eliminated, DNA is extracted, genetic diversity analysis is performed using SSR and SNP markers, the genetic distance between materials is calculated, and germplasm with a long genetic distance is collected; a collection-verification-supplementation cycle mechanism is established, and the collection focus is adjusted annually according to breeding needs; short-term preservation: the collected living materials are immediately placed in a moisturizing bag and transported back to the laboratory for tissue culture seedling induction within 24 hours; long-term preservation: after the tissue culture seedlings are confirmed to be genetically stable through more than three generations of asexual reproduction, some are moved to an intelligent incubator for long-term storage, and some are transferred to a low-temperature warehouse for backup; the morphological data, molecular marker results, and environmental parameters of the collected materials are entered into the germplasm resource database and linked to subsequent evaluation data.

3. The method for evaluating and breeding sugarcane germplasm resources for smart agriculture according to claim 1, characterized in that: The establishment of intelligent incubators and greenhouses to simulate the native environment of closely related germplasms, using IoT sensors to monitor environmental parameters, and automatically adjusting through AI algorithms to maintain the genetic stability of germplasm resources includes the following steps: Investigation of in situ environmental parameters: Restore the temperature, humidity, light, and soil environmental parameters of the native habitat of closely related germplasms to provide a regulatory basis for intelligent breeding; during the field survey phase of step one, simultaneously record the environmental parameters of the native habitat of the target germplasm; consult academic literature or international databases to supplement parameters that cannot be obtained through field surveys, including soil pH and organic matter content; classify the collected parameters by species, use statistical methods to screen key environmental factors, and establish a species-environmental parameter correspondence model; Design and build smart incubators and greenhouses: Cultivation box: suitable for small amounts of germplasm, needs to be airtight, with adjustable lighting, temperature and humidity control modules; Greenhouse: Suitable for large-scale germplasm conservation, equipped with sunshade nets, wet curtains and fans for cooling, geothermal heating pipes, and CO2 dosing equipment; Deploy IoT sensors: Temperature and humidity sensors, light sensors, and soil sensors are evenly distributed throughout the incubator and greenhouse. These sensors upload data to edge computing nodes via LoRa or ZigBee protocols, which then transmit the data to the cloud platform in real time. Develop and train AI algorithms: AI algorithms analyze sensor data and automatically adjust environmental parameters to dynamically match the in-situ model. PID control algorithms are used to set target temperatures and adjust heating and cooling power based on the difference between the current temperature and the target. Building on PID, an LSTM prediction model is introduced to combine historical environmental data to predict future environmental trends and adjust equipment in advance. In-situ survey data is used as a training set to verify the algorithm's accuracy in responding to environmental fluctuations. Maintaining genetic stability: Shoot tips and leaves of wild germplasm were collected as explants. MS culture medium composition was adjusted according to species. Explants were sterilized in a clean bench and inoculated into culture medium. Cultured in the dark at 25°C for one week, the cells were then switched to a 16-hour light / 8-hour dark cycle. Plant height, leaf shape, and stem diameter of the tissue-cultured seedlings were regularly recorded, revealing no significant differences compared to native plants. DNA from the tissue-cultured seedlings was extracted, and genetic consistency was analyzed using SSR and SNP markers. Data Management: The cloud platform displays environmental parameters and equipment status including temperature, humidity, and light in real time. Abnormal data triggers an alarm and notifies the administrator. The AI ​​algorithm analyzes sensor data every 10 minutes and automatically adjusts the equipment. The environmental model parameters are updated according to seasonal changes in the native habitat. After more than three generations of tissue culture to confirm genetic stability, the tissue culture seedlings are moved to an intelligent incubator for long-term storage, or transferred to a low-temperature warehouse for backup. The environmental parameters including temperature, humidity, and light during the cultivation process, the tissue culture seedling growth data, and genetic test results are entered into the germplasm resource database and associated with the phenotypic and genotypic data of step two to provide complete background information for subsequent evaluation.

4. The method for evaluating and breeding sugarcane germplasm resources for smart agriculture according to claim 1, wherein: The high-throughput phenotypic data collection of closely related germplasm resources using drones, robots, and near-infrared spectroscopy technology, covering agronomic traits, stress resistance, and quality indicators, includes the following steps: Preliminary preparation: Identify the phenotypic indicators and technical requirements for agronomic traits, stress resistance, and quality that need to be collected. Select a multispectral or hyperspectral drone equipped with a 20-megapixel RGB camera and a 5-band multispectral sensor, with a flight altitude of 30-50 meters. Use a field phenotyping robot equipped with a lidar, a high-precision camera, and a robotic arm capable of gripping leaves for near-infrared detection, with a travel speed set to 0.5-1 m / s. Select a portable near-infrared spectrometer with a wavelength range of 400-2500 nm and a matching leaf clamp. Agronomic traits collected: The drones scanned a wide area, designed routes based on the experimental field layout, with an overlap rate set at 70%, and flew in the early morning and evening. They collected RGB images for plant height measurement and canopy cover analysis. Multispectral images were also collected to calculate the Normalized Vegetation Index and Leaf Area Index, and assess plant growth potential. The robots performed detailed supplementary surveys, locating the areas to be supplemented based on drone images and driving along a preset path. LiDAR scanning was used to construct a 3D point cloud model of the plants, calculating stem diameter and tiller number. High-definition cameras were used to capture the images, and AI image recognition was combined to automatically identify detailed characteristics such as leaf number, leaf shape, and stem node color. Detection of stress resistance indicators: In the greenhouse, some accessions were drought-treated by withholding irrigation for 7-10 days, while a control group maintained with normal irrigation. Drone thermal infrared imaging was used to monitor leaf temperature and calculate the difference between canopy temperature and air temperature. Robotic soil sensors monitored soil moisture content in real time and correlated this with the degree of plant wilting. Target accessions were inoculated with common sugarcane pathogens, and isolation zones were set up to prevent cross-infection. Drones identified lesions using characteristic disease bands and calculated disease indexes. Robots also detected pathogen spectral signatures at the leaf level to identify latent infections at an early stage. Testing quality indicators: Mature leaves from different germplasms were collected, freeze-dried, crushed, and sieved as standard samples. These samples were scanned using a near-infrared spectrometer, and quality indicators such as sugar and fiber content were simultaneously measured. Partial least squares regression was used to establish a predictive model linking spectral data and quality indicators, with an R² validation set greater than 0.

85. A portable near-infrared spectrometer, which does not require plant damage, was used to clamp live leaves, completing spectral acquisition within 3 seconds. The data was then transmitted to a mobile app via Bluetooth, displaying predicted sugar and fiber content in real time. Quality indicators were then correlated with agronomic traits and stress resistance to screen for comprehensively superior germplasms. Data integration dynamic update: Abnormal data, including blurred drone images and robot sensor faults, is removed. Median filtering or wavelet transform is used for denoising. The data format is standardized and the cleaned data is stored by "germplasm ID - collection time - indicator type" and associated with the germplasm source information from step 1. Preliminary data analysis is performed using edge computing nodes. Abnormal data triggers an alarm, notifying researchers for on-site verification. Phenotypic data are repeatedly collected during the key growth stages of sugarcane, namely tillering, elongation, and maturity, to construct a dynamic growth curve. The near-infrared spectroscopy prediction model is regularly updated based on the newly collected data to adapt the model to different growth stages. The desensitized phenotypic data are uploaded to the National Agricultural Science Data Center for use by other research teams.

5. The method for evaluating and breeding sugarcane germplasm resources for smart agriculture according to claim 1, characterized in that: The method of combining molecular marker-assisted selection with genomic selection technology to carry out genotype identification and mine molecular markers of key genes for disease resistance, high sugar content, and stress resistance includes the following steps: Sample selection: From the closely related germplasms collected in step 1, select materials with extreme phenotypes as core samples to construct a diversity panel, including materials from different species of Saccharum and Saccharum officinale and different geographical origins, to ensure that the genetic background covers the full range of the Saccharum complex; Molecular marker development: Development of known gene markers: Disease resistance genes: refer to the literature to design PCR-based SSR or InDel markers and verify by primer specificity; High sugar gene: Using the sugarcane transporter gene sequence, SNP markers were designed and allele variation was detected by real-time fluorescence quantitative PCR; New gene marker mining: Whole-genome association analysis: Whole-genome resequencing of core samples was performed, and combined with phenotypic data, SNP markers significantly associated with target traits were screened using PLINK software; The associated SNPs were mapped to the sugarcane reference genome and compared to known functional genes by BLAST to identify candidate genes; Identify genotype: For known gene markers, sample DNA was extracted and amplified by PCR. The fragment size was read by capillary electrophoresis to determine the allele type. For high-sugar gene SNP markers, TaqMan probes were designed, and the fluorescence signal was read by a qPCR instrument to determine the genotype of the SNP site. A sugarcane-specific SNP chip was used to cover SNP sites evenly distributed throughout the genome, and genotype data were read through hybridization reactions. For samples not arrayed, GBS technology was used, sequenced on the Illumina platform, and SNP calling was performed using USTACKS software to generate a genotype matrix. Training the model: Germplasms with complete phenotypic and genotypic data were selected as the training set, and modeling was performed by grouping by traits before training. For linear models, the RR-BLUP model was used, treating the SNP effect as a random effect and solving the prediction value using a mixed linear equation. For machine learning models, random forests or support vector machines were introduced, and hyperparameters were optimized through cross-validation to improve prediction accuracy. The model's predictive ability was tested using an independent validation set, and the mean square error and r-value between the predicted and measured values ​​were calculated to improve model stability. Key gene mining: MAS marker association analysis statistically analyzes the associations between different genotypes and target traits and screens for significantly associated markers. Through SNP weight analysis using the GS model, we identify gene regions highly correlated with the target trait and identify candidate genes based on GWAS results. We overexpress or knock out candidate genes to observe whether the stress resistance of transgenic plants changes significantly. We then use CRISPR / Cas9 technology to edit candidate genes and test whether the sugar content of the edited plants decreases to confirm gene function. Data Integration: The genotype data of MAS and GS are stored in the format of germplasm ID-marker name-genotype, and the phenotypic data of step 2 are associated with the germplasm source information of step 1. In the hybrid breeding of step 3, parents carrying disease-resistant and high-sugar alleles are given priority. For the hybrid offspring, their phenotypes after adulthood are predicted using the GS model, and excellent individual plants are screened in advance to reduce field testing costs.

6. The method for evaluating and breeding sugarcane germplasm resources for smart agriculture according to claim 1, wherein: The method of constructing a hybrid combination prediction model using an AI algorithm and optimizing parental matching of closely related germplasms and cultivated sugarcane based on phenotypic and genotypic data includes the following steps: Data preprocessing: Integrate the data from steps one and two into a structured dataset as input for the AI ​​model; the data includes the agronomic traits, stress resistance, and quality indicators collected in step two; the molecular marker genotypes developed in step two, and the trait values ​​predicted by the genomic selection model; and the phenotypic and genotypic data of existing major cultivated varieties. KNN interpolation was used for phenotypic data; mean filling was used for genotypic data; Z-score normalization was performed on continuous phenotypic data; one-hot encoding was performed on genotypic data; highly redundant features were removed through correlation analysis; and the top 20 features, including drought resistance index, Bru1 marker, and GS sugar prediction value, were retained through random forest importance ranking. Define input and output: Random forests or gradient boosting trees, which are well adapted to high-dimensional data and small samples, are used. Convolutional neural networks are introduced to treat genotype data as one-dimensional sequences, and local features are extracted through convolutional layers. Input: phenotypic-genotypic joint characteristics of parental combinations; Output: Expected trait values ​​of hybrid offspring, defined as continuous or categorical values; Using historical hybridization data as the training set, parental characteristics and offspring traits were recorded. Mean squared error was used for regression tasks, and cross-entropy loss was used for classification tasks. Model parameters were adjusted through Bayesian optimization and random search, with the mean square error (MSE) and accuracy of the validation set as indicators. Multi-objective optimization: Main goal: maximize the comprehensive traits of offspring; Secondary goal: balance other traits to avoid over-optimization of a single trait; The NSGA-II optimization algorithm was selected, and parent combinations were generated through genetic operations, retaining non-dominated solutions. A penalty term was added. 100 parent combinations were randomly generated, each containing the paternal and maternal feature vectors. For each combination, the AI ​​model was used to predict offspring traits and calculate fitness values. A tournament selection method was used to retain combinations with high fitness. A single-point crossover was performed on the parent feature vectors, randomly changing parental characteristics with a 5% probability. The selection process was terminated when the fitness value did not improve for 10 consecutive generations, and the non-dominated solution set was output. Verification optimization: Using K-fold cross-validation, the data was divided into a training set and a validation set. The average MSE and accuracy of the model on the validation set were calculated to improve the model stability. The prediction accuracy of the model was tested using new hybrid combinations that had not participated in the training. The correlation coefficient between the predicted value and the measured value was calculated. Based on the test results, the model features and algorithm parameters were adjusted to improve the prediction accuracy. Hybridization experiments: Based on the optimization results, the top 20 parent combinations are output, and the expected offspring traits of each combination are annotated. A hybrid isolation zone is set up in a greenhouse or net house to prevent pollen contamination. Related germplasms are artificially emasculated and bagged with pollen from cultivated sugarcane. The hybridization date and combination are marked. After harvesting, the hybrid seeds are dried and disinfected, and sown in an intelligent incubator for germination testing. A germination rate greater than 80% is considered acceptable. Dynamic Updates: Perform phenotypic collection in step 2 on the hybrid F1 generation, compare with the model prediction value, calculate the prediction error, incorporate the newly verified hybrid combination data into the training set, retrain the model every quarter to improve the prediction ability of new combinations; incorporate the excellent hybrid offspring into the germplasm resource library in step 1, and update the genetic diversity data.

7. The method for evaluating and breeding sugarcane germplasm resources for smart agriculture according to claim 1, characterized in that: The dynamic monitoring and data collection of hybrid offspring, combined with real-time analysis on the smart agriculture platform, and screening based on preset agronomic traits, stress resistance and quality standards include the following steps: Smart agriculture platform real-time analysis: Raspberry Pi nodes deployed in the field perform preliminary cleaning and real-time analysis of sensor data, including temperature, humidity, and light. Based on Alibaba Cloud's ECS server, integrated with the Hadoop / Spark big data framework, they store comprehensive data, including images, spectra, and sensor data, and perform real-time analysis using the Flink stream processing engine. They associate agronomic traits, stress resistance, and quality indicators of each plant with a unique ID, constructing individual-level data archives. For each hybrid offspring, a growth curve model is created, and adult traits are predicted using the LSTM algorithm. Thresholds are set, triggering alarms and marking plants for retesting. Standard setting: Agronomic traits: plant height greater than 3m at maturity, number of tillers greater than 5, stem diameter greater than 2cm; Stress resistance: Drought resistance index greater than 0.8, survival rate after drought treatment greater than 80%, disease resistance rate greater than 90%, and lesion area after inoculation <10%; Quality indicators: sugar content greater than 14%, fiber content less than 12%; filter: Initial screening during the mid-growth period: During the tillering period 90 days after sowing, use drone multispectral data to screen plants with strong growth potential and no lesions, with an NDVI ≥ 0.

7. Eliminate plants with an NDVI < 0.5 or lesion area > 5%. Rescreening at maturity: During the elongation period 180 days after sowing, combining robotic lidar data with near-infrared spectroscopy data, we retain plants that meet the following criteria: plant height greater than 3 meters, stem diameter greater than 2 cm, and sugar content greater than 14%; Final screening: For plants that pass the re-screening, a 7-day drought treatment without irrigation and pathogen inoculation are carried out to eliminate individuals with CTD < 0.8 or disease resistance rate less than 90%, and finally retain the top 20% of excellent lines.

8. The method for evaluating and breeding sugarcane germplasm resources for smart agriculture according to claim 1, wherein: The above-mentioned collaborative promotion among government, enterprises and scientific research institutions, combined with technical training for farmers, will enhance the application of superior varieties, including the following steps: Clarify the responsibilities of the three parties: Establish a coordinated promotion mechanism with government guidance, enterprise leadership, and scientific research support to ensure seamless transition of superior strains from the laboratory to the field; Government: Formulate promotion policies including seedling subsidies and planting insurance, provide financial support, build promotion platforms, and coordinate cooperation between enterprises and scientific research institutions; Enterprise: Responsible for seedling production and supply, building demonstration bases, and showcasing the yield and stress resistance advantages of superior strains; Scientific research institutions: provide technical support, carry out farmer training, and solve technical problems in planting; Implementation process: First year of pilot promotion: Priority will be given to selecting 3-5 pilot counties in major sugarcane-producing areas, with each county establishing a demonstration base covering an area of ​​at least 50 mu (approximately 1.5 acres). Enterprises will produce seedlings that meet local ecological conditions based on the strain characteristics provided by scientific research institutions. Scientific research institutions will equip the demonstration bases with intelligent equipment, including IoT sensors and drones, and develop planting plans. Large-scale promotion from the second year: Through tissue culture plants or seed-stem propagation fields, the company will increase the annual production capacity of successful pilot varieties to over one million plants. Based on pilot data, the promotion area will be expanded to surrounding counties and cities, forming a core demonstration and radiating drive pattern. Build a technical training system for farmers: Training in basic planting techniques, including the characteristics of superior varieties, planting density, and fertilization periods; instruction in the use of IoT sensors, drone plant protection, and smart device operation, including intelligent water and fertilizer module control; instruction in data interpretation and decision-making, viewing planting data through a mobile app, and field management based on thresholds; deployment of soil temperature, humidity, and light sensors in promotional fields, with data transmitted to farmers' mobile apps via the LoRa network; companies providing drone plant protection services or training farmers to operate independently; installation of integrated water and fertilizer equipment based on water and fertilizer plans developed by scientific research institutions, remote control of irrigation time and fertilizer concentration through an app; deployment of pest and disease monitoring sensors, uploading data to the scientific research institution platform, identification of disease types through AI models, and recommendation of biopesticides or low-toxic chemical pesticides based on early warning information, with application times set. Effect evaluation feedback: Compare the yield, sugar content, and fiber content of superior strains with those of local dominant varieties, and calculate the rate of increase in yield and the extent of quality improvement. Compile data on drought and disease resistance in promoted plots and compare it with data from demonstration bases to verify strain stability. Investigate changes in farmers' income after planting superior strains and calculate the input-output ratio. Submit questions through the farmer app, and research institutions will respond and resolve them within 48 hours. Based on promotion data, adjust parent selection and update strains for the next year's promotion. Optimize smart devices and software based on farmer feedback. Sustainable Development: The tripartite cooperation continues: Government: Allocate special funds annually to support enterprises in expanding seedling propagation and scientific research institutions in upgrading technology; Enterprises: 10% of the promotion income will be fed back to scientific research institutions for the selection and breeding of new varieties; Scientific research institutions: launch at least two new superior strains each year and update technical solutions; Farmer participation: Farmers are encouraged to establish sugarcane planting cooperatives, purchase seedlings and intelligent equipment in a unified manner, reduce individual household costs, and allow farmers with large-scale planting and good technical skills to invest in enterprises with technology or land and share the promotion benefits.

Images