A method for monitoring crop growth and a sweet potato planting management platform

Abstract

The invention discloses a crop growth monitoring method and a sweet potato planting management platform, and relates to the technical field of agricultural management. The invention comprises the steps of obtaining stem and leaf images of different plants in different cultivation days; performing identification marking and morphological feature extraction on plants in the state of stem and leaf leggy growth, and training to obtain a leggy growth recognition model for identifying plants in different cultivation days; obtaining the cultivation days and stem and leaf images of the current plant; extracting the morphological features of the stem and leaf images of the current plant, and respectively inputting leggy growth recognition models corresponding to the cultivation days and adjacent cultivation days, to obtain several target leggy growth recognition models applicable to the current plant; judging whether the current plant is in the state of stem and leaf leggy growth according to the recognition results of the morphological features of the stem and leaf images of the current plant by each target leggy growth recognition model; and if so, sending a leggy growth processing work order. The invention can quickly and accurately identify whether the plant is in the state of stem and leaf leggy growth, and improve the efficiency of crop management.

Description

Technical Field

[0001] The invention belongs to the technical field of agricultural management, and in particular relates to a crop growth monitoring method and a sweet potato planting management platform. Background Art

[0002] With the continuous advancement of agricultural modernization, precision agriculture has become an important means to improve crop yield and quality. As an important economic crop, sweet potato is widely popular because of its rich nutrition and strong adaptability. However, the growth process of sweet potato is affected by many factors, such as climatic conditions, soil characteristics, water management, etc. These factors not only affect the growth state of sweet potato, but are also directly related to yield and quality.

[0003] At present, traditional sweet potato planting management mainly relies on farmers' experience and lacks scientific monitoring methods and data support. This method has problems such as information asymmetry and inaccurate decision-making, resulting in waste of resources and reduced yields. Therefore, there is an urgent need for a systematic monitoring method to obtain sweet potato growth information in real time and conduct scientific management. Summary of the invention

[0004] The purpose of the present invention is to provide a crop growth monitoring method and a sweet potato planting management platform, which can quickly and accurately identify whether the sweet potato stems and leaves are in a state of excessive growth by identifying the sweet potato stems and leaves during the growth period, thereby improving the crop management efficiency.

[0005] To solve the above technical problems, the present invention is achieved through the following technical solutions:

[0006] The present invention provides a crop growth monitoring method, comprising:

[0007] Obtain stem and leaf images of different plants at different cultivation days;

[0008] Identify and mark plants with elongated stems and leaves and extract morphological features, and train a model for identifying elongated plants with different cultivation days;

[0009] Get the cultivation days and stem and leaf images of the current plant;

[0010] Extracting the morphological features of the stem and leaf images of the current plant, and inputting the corresponding growth recognition models of the cultivation days and the adjacent cultivation days respectively, to obtain several target growth recognition models applicable to the current plant;

[0011] Judging whether the current plant is in a state of stem and leaf elongation according to the recognition result of the morphological features of the stem and leaf image of the current plant by each target elongation recognition model;

[0012] If yes, a stem and leaf elongation treatment work order is sent including the location of the plant in the elongated state;

[0013] If not, no processing is performed.

[0014] The present invention also discloses a crop growth monitoring method, comprising:

[0015] Receive work orders for processing excessive growth;

[0016] The crops are processed according to the positions of the plants in the state of excessive stem and leaf growth in the excessive growth processing work list.

[0017] The present invention also discloses a sweet potato planting management platform, comprising:

[0018] Image sensor, used to collect images of stems and leaves of different plants at different cultivation days;

[0019] The leggy identification terminal is used to obtain the stem and leaf images of different plants at different cultivation days;

[0020] Identify and mark plants with elongated stems and leaves and extract morphological features, and train a model for identifying elongated plants with different cultivation days;

[0021] Get the cultivation days and stem and leaf images of the current plant;

[0022] Extracting the morphological features of the stem and leaf images of the current plant, and inputting the corresponding growth recognition models of the cultivation days and the adjacent cultivation days respectively, to obtain several target growth recognition models applicable to the current plant;

[0023] Judging whether the current plant is in a state of stem and leaf elongation according to the recognition result of the morphological features of the stem and leaf image of the current plant by each target elongation recognition model;

[0024] If yes, a stem and leaf elongation treatment work order is sent including the location of the plant in the elongated state;

[0025] If not, no processing is performed;

[0026] A field management terminal for receiving excessive growth processing work orders;

[0027] The crops are processed according to the positions of the plants in the state of excessive stem and leaf growth in the excessive growth processing work list.

[0028] The present invention collects stem and leaf images of crops such as sweet potatoes through an image sensor, and then establishes and trains a leggy recognition model for identifying plants of different cultivation days through a leggy recognition terminal. During the recognition process, the leggy recognition terminal selects several target leggy recognition models suitable for the current plant and uses them to determine whether the current plant is in a state of leggy stems and leaves. This solution recognizes the stems and leaves of sweet potatoes in the growth period, and can quickly and accurately identify whether they are in a state of leggy stems and leaves, thereby improving the efficiency of crop management.

[0029] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0031] Figure 1 A schematic diagram of the functional units and information flow of a sweet potato planting management platform in one embodiment of the present invention;

[0032] Figure 2 It is a schematic diagram of the steps of an embodiment of the overgrowth identification terminal of the present invention;

[0033] Figure 3 A schematic diagram of a step flow of a field management terminal according to an embodiment of the present invention;

[0034] Figure 4 is a schematic diagram of a step flow chart of step S2 in an embodiment of the present invention;

[0035] Figure 5 is a schematic diagram of a step flow chart of step S4 in an embodiment of the present invention;

[0036] Figure 6 is a schematic diagram of a step flow chart of step S43 in an embodiment of the present invention;

[0037] Figure 7 is a schematic diagram of a step flow chart of step S434 in an embodiment of the present invention;

[0038] Figure 8 is a schematic diagram of a step flow chart of step S44 in an embodiment of the present invention;

[0039] Fig. 9 is a schematic diagram of a step flow chart of step S5 of an embodiment of the present invention;

[0040] In the accompanying drawings, the components represented by the reference numerals are listed as follows:

[0041] 1- Image sensor, 2- Excessive growth identification terminal, 3- Field management terminal. DETAILED DESCRIPTION

[0042] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0043] It should be noted that the terms "first", "second", etc. in this application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0044] See also Figures 1 to 3 As shown, the present invention provides a sweet potato planting management platform, including three interactive unit modules, namely, an image sensor 1, a leggy identification terminal 2, and a field management terminal 3. The image sensor 1 is responsible for collecting images, the leggy identification terminal 2 is responsible for identifying and judging the state of the plant, and the field management terminal 3 is a specific execution mechanism.

[0045] In the specific implementation of this solution, the image sensor 1 first executes step S101 to collect stem and leaf images of different plants at different cultivation days. Then, the leggy identification terminal 2 executes step S1 to obtain stem and leaf images of different plants at different cultivation days.

[0046] See also Figure 2 and 4As shown, before the overgrown identification terminal 2 formally identifies the state of the plant, it is necessary to train the identification model. First, step S2 can be executed to identify the plants in the overgrown state and extract the morphological features, and train to obtain the overgrown identification model for identifying plants of different cultivation days. Specifically, step S21 can be executed to mark the plants in the overgrown state to obtain the stem and leaf images of the plants in the overgrown state and the normal state with different cultivation days. Next, step S22 can be executed to extract the graphic features of the stem and leaf images of the plants in the overgrown state and the normal state with different cultivation days to obtain the morphological features of the plants in the overgrown state and the normal state with different cultivation days. Next, step S23 can be executed to establish the overgrown identification model for each cultivation day. Finally, step S24 can be executed to use the morphological characteristics of plants with different cultivation days as the input layer, and the corresponding stem and leaf elongation state or normal state as the output layer, and train the elongation recognition model for each cultivation day until convergence to obtain the elongation recognition model for identifying plants with different cultivation days.

[0047] See also Figure 2 and 5 As shown, in the process of identifying plants by the overgrowth identification terminal 2, step S3 can be first executed to obtain the cultivation days and stem and leaf images of the current plant. Next, step S4 can be executed to extract the morphological features of the stem and leaf images of the current plant, and the overgrowth identification models corresponding to the cultivation days and the adjacent cultivation days can be respectively input to obtain several target overgrowth identification models suitable for the current plant. In the specific implementation process, step S41 can be first executed to use the cultivation days corresponding to the current plant and the first several cultivation days and the last several cultivation days as reference cultivation days. Next, step S42 can be executed to input the morphological features of the stem and leaf images of the current plant into the overgrowth identification models corresponding to each reference cultivation day, and identify the probability of each output item of the output layer of the overgrowth identification model corresponding to each reference cultivation day.

[0048] See also Figure 5 and 6As shown, step S43 can be executed next to determine whether the overgrown identification model corresponding to the reference cultivation days includes all target overgrown identification models applicable to the current plant according to the probability of each output item of the output layer of the overgrown identification model corresponding to each reference cultivation day. In order to determine whether all available overgrown identification models have been selected as the target overgrown identification model, step S431 can be executed first to calculate the absolute value of the difference between the probability of the overgrown state of stems and leaves and the normal state in the output item of the output layer of the overgrown identification model corresponding to each reference cultivation day as the applicability of the overgrown identification model of each reference cultivation day. Step S432 can be executed next to determine the numerical range of the applicability of the target overgrown identification model applicable to the current plant according to the probability distribution of the output items of the overgrown identification model of each reference cultivation day and the applicability as the target applicability range. Step S433 can be executed next to take the reference cultivation days whose applicability of the corresponding overgrown identification model falls within the target applicability range as the target cultivation days. Next, step S434 may be executed to determine whether the date range of the target cultivation days is at one end of the date range of all current reference cultivation days. If so, step S435 may be triggered to determine whether the leggy identification model corresponding to the current reference cultivation days does not include all target leggy identification models applicable to the current plant. If not, step S436 may be triggered to determine whether the leggy identification model corresponding to the current reference cultivation days includes all target leggy identification models applicable to the current plant.

[0049] In order to supplement the implementation process of the above steps S431 to S436, the source code of some functional modules is provided, and the explanation is compared in the comment section. In order to avoid the leakage of data involving commercial secrets, some data that does not affect the implementation of the solution are desensitized, the same below.

[0050] #include <iostream>

[0051] #include <vector>

[0052] #include <map>

[0053] #include<opencv2 / opencv.hpp> / / Introduce OpenCV library

[0054] #include<opencv2 / ml.hpp> / / Introduce machine learning module

[0055] / / Define the overgrowth recognition model class

[0056] class ElongationModel {

[0057] public:

[0058] cv::Ptr <cv::ml::knearest>model; / / K nearest neighbor model

[0059] ElongationModel() {

[0060] model = cv::ml::KNearest::create(); / / Initialize K nearest neighbor model

[0061] }

[0062] / / How to train the model

[0063] void train(const cv::Mat&trainingData, const cv::Mat&labels) {

[0064] model->train(trainingData, cv::ml::ROW_SAMPLE, labels);

[0065] }

[0066] / / Method for predicting the state of leggy growth

[0067] cv::Mat predictProbabilities(const cv::Mat&features) {

[0068] cv::Mat results;

[0069] model->findNearest(features, 3, results);

[0070] return results; / / Return prediction results

[0071] }

[0072] };

[0073] / / Functions for calculating suitability and judging suitability

[0074] void evaluateModelSuitability(const std::map<int, ElongationModel> &models, const cv::Mat&featureMat, const std::vector <int>&referenceDays,double targetSuitabilityRangeLow, double targetSuitabilityRangeHigh) {

[0075] std::vector <int>suitableDays; / / Store suitable cultivation days

[0076] std::map<int, double> suitabilityScores; / / Store suitability scores

[0077] / / 1. Traverse each reference cultivation day and calculate the suitability

[0078] for (int day : referenceDays) {

[0079] if (models.find(day) != models.end()) { / / Check if the model exists

[0080] cv::Mat probabilities = models.at(day).predictProbabilities(featureMat);

[0081] double elongationProbability = probabilities.at <float>(0, 1); / / Probability of elongated state

[0082] double normalProbability = probabilities.at <float>(0, 0); / / Probability of normal state

[0083] / / Calculate suitability

[0084] double suitabilityScore = std::abs(elongationProbability -normalProbability);

[0085] suitabilityScores[day] = suitabilityScore; / / Store suitability scores

[0086] / / Judge whether the applicability falls within the target range

[0087] if (suitabilityScore>= targetSuitabilityRangeLow&&suitabilityScore<=targetSuitabilityRangeHigh) {

[0088] suitableDays.push_back(day); / / Record the suitable cultivation days

[0089] }

[0090] }

[0091] }

[0092] / / 2. Determine whether the target cultivation days date range is at one end of the reference cultivation days date range

[0093] if (!suitableDays.empty()) {

[0094] int minDay = *std::min_element(suitableDays.begin(), suitableDays.end());

[0095] int maxDay = *std::max_element(suitableDays.begin(), suitableDays.end());

[0096] if (minDay<referenceDays.front() || maxDay> referenceDays.back()) {

[0097] std::cout<<"The leggy identification model corresponding to the current reference cultivation days does not include all target leggy identification models applicable to the current plant."< <std::endl;

[0098] } else {

[0099] std::cout<<"The leggy identification model corresponding to the current reference cultivation days includes all target leggy identification models applicable to the current plant."< <std::endl;

[0100] }

[0101] } else {

[0102] std::cout<<"No applicable target elongation identification model was found."< <std::endl;

[0103] }

[0104] }

[0105] int main() {

[0106] / / Example: Load the stem and leaf images of the current plant

[0107] cv::Mat currentImage = cv::imread("current_plant_image.jpg");

[0108] int currentDay = 10; / / Example: the current cultivation days are 10 days

[0109] / / Example: Preload the leggy identification model for different cultivation days

[0110] std::map<int, ElongationModel> models; / / Store models for different cultivation days

[0111] / / TODO: Load model logic

[0112] / / Extract the morphological characteristics of the current plant

[0113] FeatureExtractor extractor;

[0114] std::vector <double>currentFeatures = extractor.extractFeatures(currentImage);

[0115] cv::Mat featureMat(currentFeatures, true); / / Convert to cv::Mat

[0116] / / Set the target applicability range

[0117] double targetSuitabilityRangeLow = 0.1; / / Example suitability range lower limit

[0118] double targetSuitabilityRangeHigh = 0.5; / / Example suitability range upper limit

[0119] / / Evaluate suitability

[0120] std::vector <int>referenceDays = {8, 9, 10, 11, 12}; / / Example reference cultivation days

[0121] evaluateModelSuitability(models, featureMat, referenceDays, targetSuitabilityRangeLow, targetSuitabilityRangeHigh);

[0122] return 0;

[0123] }

[0124] The code implements the process of judging whether the target leggy identification model applicable to the current plant is included according to the output probability of the leggy identification model corresponding to each reference cultivation day. The main steps include calculating the applicability of each reference cultivation day model, judging whether the applicability is within the target range, recording the applicable cultivation days, and judging whether the target model of the current plant is included according to the date range of these cultivation days. There is also a TODO part in the code for the loading logic of the model.

[0125] See also Figure 7 As shown, in the judgment logic of the above step S434, it is necessary to find multiple target overgrowth identification models suitable for the current plant. These target overgrowth identification models have commonalities in the output item probability distribution and the numerical values ​​of the applicability, but they are not completely consistent. In order to quantify this commonality and divide the range, firstly, step S4341 can be executed to select the reference cultivation days corresponding to the maximum and minimum values ​​in all applicability as the benchmark reference cultivation days. Next, step S4342 can be executed to calculate the cumulative value of the output item probability of the stem and leaf overgrowth state, the normal state output item probability and the absolute value of the applicability difference of the overgrowth identification model of each benchmark reference cultivation day and other reference cultivation days as the recognition difference between each benchmark reference cultivation day and the overgrowth identification model of other reference cultivation days. Next, step S4343 can be executed for each reference cultivation day other than the benchmark reference cultivation day, and each other reference cultivation day and the benchmark reference cultivation day with the smallest recognition difference are classified into the same reference cultivation day combination. Next, step S4344 may be executed to calculate and obtain the mean value of the fitness of the excessive growth identification model for each reference cultivation day in each reference cultivation day combination.

[0126] In order to determine whether the current reference cultivation days combination is appropriate, step S4345 can be executed next to determine whether each reference cultivation days in the reference cultivation days combination with a larger mean value is adjacent and coherent. If so, step S4346 can be executed next to use the range of applicability of the leggy identification model of each reference cultivation days in the reference cultivation days combination with a larger mean value as the target applicability range. If not, it means that the current reference cultivation days combination is not appropriate, so steps S4341 to S4345 can be executed next to change the benchmark reference cultivation days, recalculate the classification reference cultivation days group, and merge for judgment.

[0127] In order to provide additional explanation for the implementation process of the above-mentioned steps S4341 to S4346, source code of some functional modules is provided, and comparative explanations are provided in the comment section.

[0128] #include <iostream>

[0129] #include <vector>

[0130] #include <map>

[0131] #include <algorithm>

[0132] #include<opencv2 / opencv.hpp> / / Introduce OpenCV library

[0133] #include<opencv2 / ml.hpp> / / Introduce machine learning module

[0134] / / Define the overgrowth recognition model class

[0135] class ElongationModel {

[0136] public:

[0137] cv::Ptr <cv::ml::knearest>model; / / K nearest neighbor model

[0138] ElongationModel() {

[0139] model = cv::ml::KNearest::create(); / / Initialize K nearest neighbor model

[0140] }

[0141] / / How to train the model

[0142] void train(const cv::Mat&trainingData, const cv::Mat&labels) {

[0143] model->train(trainingData, cv::ml::ROW_SAMPLE, labels);

[0144] }

[0145] / / Method for predicting the state of leggy growth

[0146] cv::Mat predictProbabilities(const cv::Mat&features) {

[0147] cv::Mat results;

[0148] model->findNearest(features, 3, results);

[0149] return results; / / Return prediction results

[0150] }

[0151] };

[0152] / / Calculate the applicable range of reference cultivation days

[0153] void computeTargetSuitabilityRange(const std::map<int,ElongationModel> &models, const std::vector <int>&referenceDays) {

[0154] std::map<int, double> suitabilityScores; / / Store suitability scores

[0155] / / 1. Calculate the suitability of each reference cultivation day

[0156] for (int day : referenceDays) {

[0157] if (models.find(day) != models.end()) {

[0158] cv::Mat probabilities = models.at(day).predictProbabilities(cv::Mat()); / / Test with empty features

[0159] double elongationProbability = probabilities.at <float>(0, 1); / / Probability of elongated state

[0160] double normalProbability = probabilities.at <float>(0, 0); / / Probability of normal state

[0161] double suitabilityScore = std::abs(elongationProbability -normalProbability);

[0162] suitabilityScores[day] = suitabilityScore; / / Store suitability scores

[0163] }

[0164] }

[0165] / / 2. Select the maximum and minimum values ​​of suitability as the reference cultivation days

[0166] auto maxIt = std::max_element(suitabilityScores.begin(),suitabilityScores.end(),

[0167] [](const auto&a, const auto&b) { return a.second <b.second;});

[0168] auto minIt = std::min_element(suitabilityScores.begin(),suitabilityScores.end(),

[0169] [](const auto&a, const auto&b) { return a.second <b.second;});

[0170] int baseDayMax = maxIt->first;

[0171] int baseDayMin = minIt->first;

[0172] / / 3. Calculate the recognition difference

[0173] std::map<int, double> recognitionDifferences; / / Store recognition differences

[0174] for (int day : referenceDays) {

[0175] if (day != baseDayMax&&day != baseDayMin) {

[0176] double difference = std::abs(suitabilityScores[baseDayMax] -suitabilityScores[day]) +

[0177] std::abs(suitabilityScores[baseDayMin] - suitabilityScores[day]);

[0178] recognitionDifferences[day] = difference; / / Store the difference

[0179] }

[0180] }

[0181] / / 4. Classification reference cultivation days combination

[0182] std::map <int, std::vector <int>>dayGroups; / / Store reference cultivation day combinations

[0183] for (const auto&entry : recognitionDifferences) {

[0184] int day = entry.first;

[0185] / / Classify the reference cultivation days with the minimum recognition difference

[0186] int closestBaseDay = (recognitionDifferences[baseDayMax] <recognitionDifferences[baseDayMin]) ? baseDayMax : baseDayMin;

[0187] dayGroups[closestBaseDay].push_back(day);

[0188] }

[0189] / / 5. Calculate the mean and determine adjacency

[0190] for (const auto&group : dayGroups) {

[0191] int baseDay = group.first;

[0192] const auto&daysInGroup = group.second;

[0193] double totalSuitability = suitabilityScores[baseDay];

[0194] double count = 1; / / Includes the base day

[0195] for (int day : daysInGroup) {

[0196] totalSuitability += suitabilityScores[day];

[0197] count++;

[0198] }

[0199] double meanSuitability = totalSuitability / count;

[0200] / / Judge whether the days in the combination with the larger mean are adjacent

[0201] std::vector <int>sortedDays = daysInGroup;

[0202] sortedDays.push_back(baseDay);

[0203] std::sort(sortedDays.begin(), sortedDays.end());

[0204] bool isConsecutive = true;

[0205] for (size_t i = 1; i < sortedDays.size(); ++i) {

[0206] if (sortedDays[i] != sortedDays[i - 1]+ 1) {

[0207] isConsecutive = false;

[0208] break;

[0209] }

[0210] }

[0211] / / Output result

[0212] if (isConsecutive) {

[0213] std::cout << "The target suitability range is: " << meanSuitability << " (Reference cultivation days: " << baseDay << ")" << std::endl;

[0214] } else {

[0215] std::cout << "The combination of reference cultivation days " << baseDay << " is not consecutive. Recalculate." << std::endl;

[0216] / / The logic to continue changing the reference cultivation days and recalculating can be added here

[0217] }

[0218] }

[0219] }

[0220] int main() {

[0221] / / Example: Preload the leggy identification model for different cultivation days

[0222] std::map<int, ElongationModel> models; / / Store models for different cultivation days

[0223] / / TODO: Load model logic

[0224] / / Example reference cultivation days

[0225] std::vector <int>referenceDays = {8, 9, 10, 11, 12};

[0226] computeTargetSuitabilityRange(models, referenceDays);

[0227] return 0;

[0228] }

[0229] The code implements the process of judging the suitability range of the target leggy identification model for the current plant based on the probability distribution of the leggy identification model output for each reference cultivation day. The main steps include selecting the maximum and minimum values ​​of the suitability as the benchmark reference cultivation days, calculating the recognition difference, classifying the reference cultivation days combination, and calculating the mean and judging the adjacency to determine whether to use the suitability range of the combination with the larger mean as the target range. There is a TODO part in the code for the loading logic of the model.

[0230] Please continue to refer to Figure 5 As shown, if the judgment in the above step S43 is yes, then step S44 can be executed to obtain several target leggy identification models applicable to the current plant, that is, the leggy identification models corresponding to the target cultivation days are used as several target leggy identification models applicable to the current plant. If not, it means that some leggy identification models have not been selected as target leggy identification models at this time, which may lead to the problem that the stem and leaf state recognition may not be comprehensive enough, so steps S41 to S43 can be executed to change the reference cultivation days and re-identify and judge.

[0231] See also Figure 8 As shown, since the cultivation days of sweet potatoes are long, in order to improve the recognition speed, therefore, changing the reference cultivation days should also be directional and targeted. Therefore, in the specific implementation process of the above-mentioned step S44, step S441 can be first executed to distribute the date range of the target cultivation days to one end of the current reference cultivation days date range as an extension end. Next, step S442 can be executed to determine whether the cultivation days corresponding to the extension end are the cultivation days of the current plant. If so, step S443 can be executed next without changing the reference cultivation days, and the identification and judgment are not re-performed. If not, step S444 can be executed next to use several non-reference cultivation days that are adjacent and coherent to the cultivation days corresponding to the extension end as newly added reference cultivation days, and the identification and judgment are re-performed.

[0232] See also Figure 2 and 9 As shown, in order to improve the accuracy of recognition in the above steps, multiple target overgrowth recognition models are adopted. In order to integrate multiple recognition results, first, step S5 can be executed to determine whether the current plant is in the state of overgrowth of stems and leaves according to the recognition result of the morphological characteristics of the stem and leaf images of the current plant by each target overgrowth recognition model. Next, step S51 can be executed to perform weighted accumulation according to the ratio between the fitness of each target overgrowth recognition model for the output item probability of the overgrowth state of stems and leaves and the output item probability of the normal state in the recognition output layer of the morphological characteristics of the stem and leaf images of the current plant, and obtain the weighted cumulative probability of the output item of the overgrowth state of stems and leaves and the output item of the normal state respectively. Next, step S52 can be executed to determine whether the output item with a larger weighted cumulative probability is the state of overgrowth of stems and leaves. If so, then step S53 can be triggered to determine that the current plant is in the state of overgrowth of stems and leaves. If otherwise, step S54 can be triggered to determine that the current plant is not in the state of overgrowth of stems and leaves. The recognition results of multiple target overgrowth recognition models are combined by calculating the weighted mean to improve the accuracy and comprehensiveness of recognition.

[0233] Please continue to refer to Figures 1 to 3 As shown, after executing the judgment logic in step S5, if the judgment result is yes, then step S6 can be executed to send a leggy processing work order containing the location of the plant in the leggy state. If the judgment result is no, then step S7 can be executed to not process and continue to monitor the crops.

[0234] The field management terminal 3 may be a handheld terminal of a field agricultural worker, and during the work process, step S031 may be executed to receive a superfluous processing work order, and then step S032 may be executed to process the crops according to the positions of the plants in the superfluous stem and leaf state in the superfluous processing work order.

[0235] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the device, system, method and computer program product according to multiple embodiments of the present application. In this regard, each square frame in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the logical function of the specification. In some alternative implementations, the functions marked in the square frame can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square frames can actually be executed substantially in parallel, and they can also be executed in reverse order sometimes, depending on the functions involved.

[0236] It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented by hardware that performs the corresponding function or action, such as a circuit or ASIC (Application Specific Integrated Circuit), or can be implemented by a combination of hardware and software, such as firmware.

[0237] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art may understand and implement other variations of the disclosed embodiments by viewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple situations. A single processor or other unit may implement several functions listed in the claims. Certain measures are recorded in different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0238] The embodiments of the present application have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes are obvious to those of ordinary skill in the art without departing from the scope of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.< / int> < / int> < / int> < / float> < / float> < / int> < / cv::ml::knearest> < / algorithm> < / map> < / vector> < / iostream> < / int> < / double> < / float> < / float> < / int> < / int> < / cv::ml::knearest> < / map> < / vector> < / iostream>

Claims

1. A method for monitoring crop growth, characterized in that: include, Obtain stem and leaf images of different plants at different cultivation days; Identify and mark plants with elongated stems and leaves and extract morphological features, and train a model for identifying elongated plants with different cultivation days; Get the cultivation days and stem and leaf images of the current plant; Extracting the morphological features of the stem and leaf images of the current plant, and inputting the corresponding growth recognition models of the cultivation days and the adjacent cultivation days respectively, to obtain several target growth recognition models applicable to the current plant; Judging whether the current plant is in a state of stem and leaf elongation according to the recognition result of the morphological features of the stem and leaf image of the current plant by each target elongation recognition model; If yes, a stem and leaf elongation treatment work order is sent including the location of the plant in the elongated state; If not, no processing is performed; The step of extracting the morphological features of the stem and leaf images of the current plant and inputting the corresponding growth recognition models of the cultivation days and the adjacent cultivation days to obtain several target growth recognition models suitable for the current plant includes: The cultivation days corresponding to the current plant, the previous cultivation days, and the next cultivation days are used as reference cultivation days; The morphological features of the stem and leaf images of the current plant are respectively input into the leggy identification model corresponding to each reference cultivation day, and the probability of each output item of the output layer of the leggy identification model corresponding to each reference cultivation day is identified; Judging whether the leggy identification model corresponding to each reference cultivation day includes all target leggy identification models applicable to the current plant according to the probability of each output item of the output layer of the leggy identification model corresponding to each reference cultivation day; If so, several target leggy identification models applicable to the current plant are obtained; If not, change the reference cultivation days and re-identify and judge.

2. The method according to claim 1, characterized in that: The steps of identifying and marking plants in a state of excessive stem and leaf growth, extracting morphological features, and training to obtain an excessive growth recognition model for identifying plants of different cultivation days include: The plants in the state of excessive stem and leaf growth and the normal state are marked to obtain the stem and leaf images of the plants in the state of excessive stem and leaf growth and the normal state at different cultivation days; The morphological characteristics of the plants in the state of excessive stem and leaf growth and in the state of normal cultivation for different cultivation days are respectively extracted by performing graphic feature extraction on the stem and leaf images of the plants in the state of excessive stem and leaf growth and in the state of normal cultivation for different cultivation days; Establish a model for identifying leggy growth at each cultivation day; The morphological characteristics of plants with different cultivation days are used as the input layer, and the corresponding stem and leaf elongation state or normal state is used as the output layer. The elongation recognition model for each cultivation day is trained until convergence to obtain a elongation recognition model for identifying plants with different cultivation days.

3. The method according to claim 1, characterized in that The step of judging whether the leggy identification model corresponding to each reference cultivation day includes all target leggy identification models applicable to the current plant according to the probability of each output item of the output layer of the leggy identification model corresponding to each reference cultivation day, include, For each reference cultivation day, the absolute value of the difference between the probability of the stem and leaf elongation state and the normal state in the output item of the output layer of the corresponding elongation recognition model is calculated as the applicability of the elongation recognition model for each reference cultivation day; According to the probability distribution of the output items of the leggy identification model of each reference cultivation day and the applicability judgment, a numerical range of the applicability of the target leggy identification model applicable to the current plant is obtained as a target applicability range; The reference cultivation days for which the fitness of the corresponding leggy identification model falls within the target fitness range are taken as the target cultivation days; Determine whether the date range of the target cultivation days is at one end of the date range of all current reference cultivation days; If so, it is determined that the leggy identification model corresponding to the current reference cultivation days does not include all target leggy identification models applicable to the current plant; If not, it is determined that the leggy growth recognition model corresponding to the current reference cultivation days includes all target leggy growth recognition models applicable to the current plant.

4. The method according to claim 3, characterized in that The step of obtaining the numerical range of suitability of the target overgrowth identification model applicable to the current plant as the target suitability range according to the probability distribution of the output items of the overgrowth identification model of each reference cultivation day and the suitability judgment includes: Select the reference cultivation days corresponding to the maximum and minimum values ​​of all suitability as the benchmark reference cultivation days; Calculate and obtain the cumulative value of the output item probability of the stem and leaf elongation state, the normal state output item probability and the absolute value of the fitness difference of the elongation recognition model between each benchmark reference cultivation day and other reference cultivation days as the recognition difference degree of the elongation recognition model between each benchmark reference cultivation day and other reference cultivation days; For each reference cultivation day other than the benchmark reference cultivation day, classify each other reference cultivation day and the benchmark reference cultivation day with the smallest recognition difference into the same reference cultivation day combination; Calculate and obtain the mean value of the fitness of the leggy identification model for each reference cultivation day in each reference cultivation day combination; Determine whether each reference cultivation day in the reference cultivation day combination with a larger mean is adjacent and coherent; If so, the range of applicability of the leggy identification model for each reference cultivation day in the reference cultivation day combination with a larger mean is taken as the target applicability range; If not, the benchmark reference cultivation days are changed and the classified reference cultivation days groups are recalculated and combined for judgment.

5. The method according to claim 3, characterized in that: The step of obtaining several target leggy identification models applicable to the current plant, include, The leggy identification models corresponding to the target cultivation days are used as several target leggy identification models applicable to the current plants.

6. The method according to claim 3, characterized in that The step of changing the reference cultivation days and re-identifying and judging comprises: The date range of the target cultivation days is distributed to one end of the current date range of all reference cultivation days as the extension end; Determine whether the cultivation days corresponding to the extension end are the cultivation days of the current plant; If yes, the reference cultivation days will not be changed, and the identification judgment will not be re-made; If not, several adjacent and continuous non-reference cultivation days corresponding to the extended end are used as newly added reference cultivation days, and the identification and judgment are performed again.

7. The method according to claim 3, characterized in that The step of judging whether the current plant is in a state of stem and leaf elongation according to the recognition result of the morphological features of the stem and leaf image of the current plant by each target elongation recognition model, include, For each target overgrowth recognition model, the probability of the stem and leaf overgrowth state output item and the probability of the normal state output item in the recognition output layer of the morphological characteristics of the stem and leaf image of the current plant are weighted and accumulated according to the ratio between the fitness of each target overgrowth recognition model, and the weighted accumulated probabilities of the stem and leaf overgrowth state output item and the normal state output item are obtained respectively; Determine whether the output item with a larger weighted cumulative probability is in the state of excessive stem and leaf growth; If so, the current plant is in a state of excessive stem and leaf growth; If not, the current plant is not in a state of excessive stem and leaf growth.

8. A field management terminal, characterized in that: include, Receiving a work order for processing excessive growth in a crop growth monitoring method according to any one of claims 1 to 7; The crops are processed according to the positions of the plants in the state of excessive stem and leaf growth in the excessive growth processing work list.

9. A sweet potato planting management platform, characterized in that: include, Image sensor, used to collect images of stems and leaves of different plants at different cultivation days; The leggy identification terminal is used to obtain the stem and leaf images of different plants at different cultivation days; Identify and mark plants with elongated stems and leaves and extract morphological features, and train a model for identifying elongated plants with different cultivation days; Get the cultivation days and stem and leaf images of the current plant; Extracting the morphological features of the stem and leaf images of the current plant, and inputting the corresponding growth recognition models of the cultivation days and the adjacent cultivation days respectively, to obtain several target growth recognition models applicable to the current plant; Judging whether the current plant is in a state of stem and leaf elongation according to the recognition result of the morphological features of the stem and leaf image of the current plant by each target elongation recognition model; If yes, a stem and leaf elongation treatment work order is sent including the location of the plant in the elongated state; If not, no processing is performed; The field management terminal is used to receive work orders for excessive growth treatment; Processing the crops according to the positions of the plants in the state of excessive stem and leaf growth in the excessive growth processing work order; The step of extracting the morphological features of the stem and leaf images of the current plant and inputting the corresponding growth recognition models of the cultivation days and the adjacent cultivation days to obtain several target growth recognition models suitable for the current plant includes: The cultivation days corresponding to the current plant, the previous cultivation days, and the next cultivation days are used as reference cultivation days; The morphological features of the stem and leaf images of the current plant are respectively input into the leggy identification model corresponding to each reference cultivation day, and the probability of each output item of the output layer of the leggy identification model corresponding to each reference cultivation day is identified; Judging whether the leggy identification model corresponding to each reference cultivation day includes all target leggy identification models applicable to the current plant according to the probability of each output item of the output layer of the leggy identification model corresponding to each reference cultivation day; If so, several target leggy identification models applicable to the current plant are obtained; If not, change the reference cultivation days and re-identify and judge.

Images