Agricultural product system man-machine interaction system based on Internet of Things

Abstract

The invention relates to the technical field of man-machine interaction, in particular to an agricultural product system man-machine interaction system based on the Internet of Things. The system comprises a data acquisition module, a data transmission module, a model construction module and a screen display module. According to the invention, the soil humidity, temperature and illumination intensity of the agricultural product growth environment are acquired through the data acquisition module, and the weights of the sensors of the same type and different precisions are dynamically adjusted according to the real-time data quality and historical data of the sensors, so that the accuracy of the acquired data is improved; a neural network model of the agricultural product growth condition is constructed through a model construction module, different categories or degrees of the agricultural product growth condition are judged by using the model, and a corresponding adjustment strategy is formulated according to an output result, so that the growth environment of the agricultural product is adjusted in time; normal growth of agricultural products is guaranteed, and income is increased.

Description

Technical Field

[0001] The present invention relates to the technical field of human-computer interaction, and more specifically, to a human-computer interaction system for an agricultural product system based on the Internet of Things. Background Art

[0002] The human-computer interaction system for an agricultural product system based on the Internet of Things aims to connect all aspects of agricultural production and management through Internet of Things technology, providing an efficient and intuitive human-computer interaction interface to promote the intelligence and informatization of agricultural production. It mainly includes: front-end devices, including various sensors (such as temperature and humidity, light, etc.), actuators (such as an automatic irrigation system), and mobile terminals (smartphones, tablets); a data transmission network, using wireless communication technologies such as Wi-Fi, LoRa, NB-IoT, etc. to achieve remote data transmission; a back-end service platform, including a data processing center, a cloud computing platform, and a database, for storing and analyzing the collected data; a user interface, providing a graphical interface that enables users to easily view data, receive notifications, and control devices. The front-end devices collect farmland environmental data and transmit it to the back-end service platform through the network. The back-end service platform processes and analyzes the data, generates management suggestions or warning information, and users access the system through the mobile terminal to view data, receive notifications, and control the front-end devices.

[0003] Currently, there are multiple accuracy types for the same type of sensor. For sensors used to detect the growth environment of external agricultural products, the higher the accuracy, the more easily it is affected by the external environment (such as weather, electromagnetic interference), resulting in inaccurate data collection. If the accuracy is too low, it cannot achieve the purpose of accurate measurement. Moreover, the construction of network infrastructure in rural areas is relatively backward, the network coverage is limited, and the signal quality is unstable, affecting the reliability of data transmission. Therefore, in order to improve the accuracy of measurement data, predict the growth state of agricultural products based on the measurement data, and formulate relevant adjustment strategies to ensure the good growth state of agricultural products and improve productivity, we propose a human-computer interaction system for an agricultural product system based on the Internet of Things. Summary of the Invention

[0004] The purpose of the present invention is to solve the problems that when sensors detect the growth environment of agricultural products, they are easily affected by the external environment, resulting in inaccurate data collection and unstable data transmission, and cannot timely reflect the growth status of agricultural products. In order to improve the accuracy of measurement data, predict the growth state of agricultural products based on the measurement data, and formulate relevant adjustment strategies.

[0005] To achieve the above object, the present invention provides a human-computer interaction system for an agricultural product system based on the Internet of Things, including a data acquisition module, a data transmission module, a model construction module, and a screen display module;

[0006] The data acquisition module uses multiple sensors of the same type but different precisions to collect the soil humidity, temperature, and light intensity of the agricultural product growth environment, and uses a movable high-definition camera to take pictures of the external characteristics of the agricultural products, collecting the leaf color characteristic data, shape characteristic data, and leaf spot area data of the agricultural products. After organizing and packaging the collected data, it is sent to the data transmission module;

[0007] The data transmission module uses a variety of wireless communication technologies for data transmission, mainly using 5G mobile communication technology, supplemented by Wi-Fi and Zigbee, adopting a hybrid backup mode. A Wi-Fi hotspot is set up locally as a backup. When the 5G signal is unstable, the device can automatically switch to the Wi-Fi hotspot for data transmission. Between the device and some local short-range sensors, Zigbee can be used for communication, and the collected data is transmitted to the model construction module;

[0008] The model construction module establishes an input layer with the soil humidity, temperature, light intensity, leaf color characteristic data, shape characteristic data, and leaf spot area data as input nodes, sets multiple hidden layers, each layer contains neurons, and the neurons learn the complex non-linear relationship between the input features through the ReLU activation function. Through a large amount of training data, the neural network can automatically adjust the weights between the hidden layer neurons. The output layer sets multiple nodes to represent different categories of plant growth conditions, and the output results are presented in the form of probabilities. According to the output results, corresponding adjustment strategies are formulated, and the output results and the corresponding adjustment strategies are passed to the screen display module;

[0009] The screen display module displays the corresponding image information according to the output results. At the same time, it uses voice broadcast to play the output results and adjustment strategies to notify the staff to make corresponding adjustments.

[0010] As a further improvement of this technical solution, the model construction module includes a model creation unit and an adjustment strategy unit;

[0011] The model creation unit takes the soil humidity, temperature, light intensity, leaf color characteristic data, shape characteristic data, and leaf spot area data as input, and takes different categories of plant growth conditions as output to establish a neural network model, and divides the organized data into a training set, a validation set, and a test set according to a certain proportion to train the neural network model;

[0012] The adjustment strategy unit selects the corresponding adjustment method according to the multi-node agricultural product growth conditions output by the neural network model, and selects the corresponding adjustment amplitude according to the probability of the output results.

[0013] As a further improvement of this technical solution, when training the neural network model, the model creation unit measures the difference between the predicted probability distribution and the true probability distribution of the model using the multi-class cross-entropy loss function, and trains the neural network model with the minimum loss function as the objective. The calculation formula of the multi-class cross-entropy loss function is as follows:

[0014]

[0015] where y p is the probability distribution output by the model, y t is the probability distribution of the true label, and C is the number of classes.

[0016] As a further improvement of this technical solution, when training the neural network model, the model creation unit combines the advantages of the momentum method and the adaptive learning rate using the Adam optimization algorithm, calculates the first-order moment estimate and the second-order moment estimate of the gradient, and then adjusts the learning rate and updates the parameters according to these two estimates.

[0017] As a further improvement of this technical solution, the adjustment strategy unit adopts a risk assessment-based method, combines the problem occurrence probability output by the neural network model with the harm degree, determines the priorities of different categories of growth conditions, and preferentially adopts the corresponding adjustment strategies.

[0018] As a further improvement of this technical solution, the data acquisition module includes a sensor acquisition unit and an image acquisition unit;

[0019] The sensor acquisition unit uses multiple sensors of the same type but different precisions to collect the soil humidity, temperature, and light intensity of the agricultural product growth environment;

[0020] The image acquisition unit uses a movable high-definition camera to take pictures of the appearance features of agricultural products, and extracts the leaf color feature data, shape feature data, and leaf spot area data of agricultural products.

[0021] As a further improvement of this technical solution, the sensor acquisition unit adopts the Kalman filter data fusion algorithm to dynamically adjust the weights of sensors of the same type but different precisions according to the real-time data quality and historical data of the sensors.

[0022] As a further improvement of this technical solution, the image acquisition unit extracts the leaf color feature data by calculating the mean and variance of the color channels of the leaf area, uses the edge detection algorithm to extract the edge contour of the leaf, and then extracts the shape feature data, and uses the threshold segmentation method to extract the leaf spot area data.

[0023] As a further improvement of this technical solution, the data transmission module establishes a regular detection mechanism for the communication link. By sending test data packets and monitoring the link status, it determines whether the main communication technology is working properly. If it is not in a normal working state, the communication link is switched.

[0024] As a further improvement of this technical solution, the screen display module includes an image display unit and a voice broadcast unit;

[0025] The image display unit displays the output result of the neural network and the corresponding adjustment strategy in the form of a moving picture, which is displayed in the middle of the screen;

[0026] The voice broadcast unit broadcasts the output result of the neural network and the corresponding adjustment strategy to notify the staff to make corresponding adjustments. At the same time, it receives the voice commands of the staff and, according to the commands, selects to play the key measures in the adjustment strategy.

[0027] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0028] 1. This human-computer interaction system for the agricultural product system based on the Internet of Things uses the data acquisition module to collect the soil humidity, temperature, and light intensity of the agricultural product growth environment by using multiple sensors of the same type but different precisions. The Kalman filter data fusion algorithm is adopted to dynamically adjust the weights of sensors of the same type but different precisions according to the real-time data quality and historical data of the sensors. Multiple sensors jointly collect the growth environment data of agricultural products, improving the accuracy of the collected data and reducing the influence of environmental factors on the data collected by the sensors;

[0029] 2. The data transmission module uses a variety of wireless communication technologies for data transmission, mainly 5G mobile communication technology, supplemented by Wi-Fi and Zigbee, and adopts a hybrid backup mode to ensure that the data transmission module is always in a normal working state, ensuring the stability of data transmission and improving the reliability of data transmission;

[0030] 3. The model construction module constructs a neural network model for the growth status of agricultural products, uses this model to judge different categories or degrees of the growth status of agricultural products, and formulates corresponding adjustment strategies according to the output results, timely adjusts the growth environment of agricultural products, ensures that agricultural products can grow normally, and increases the income. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0032] Figure 2 It is a schematic diagram of the overall detailed process of the present invention;

[0033] Figure 3 It is a schematic diagram of the structure of the screen display module of the present invention.

[0034] The meanings of the reference numerals in the figure are as follows:

[0035] 100, data acquisition module; 110, sensor acquisition unit; 120, image acquisition unit; 200, data transmission module; 300, model construction module; 310, model creation unit; 320, adjustment strategy unit; 400, screen display module; 410, image display unit; 420, voice broadcast unit. Detailed implementation manners

[0036] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0037] Currently, when sensors detect the growth environment of agricultural products, they are easily affected by the external environment, resulting in inaccurate collected data and unstable data transmission, and cannot timely reflect the growth status of agricultural products. In order to improve the accuracy of measurement data, predict the growth state of agricultural products based on the measurement data, and formulate relevant adjustment strategies.

[0038] Therefore, the present invention proposes to collect the growth environment data of agricultural products through the data acquisition module using multiple sensors of the same type but different precisions, use the data transmission module to perform data transmission using multiple wireless communication technologies, and the model construction module constructs a neural network model of the growth status of agricultural products, and uses the model to judge different categories or degrees of the growth status of agricultural products, and formulate corresponding adjustment strategies according to the output results.

[0039] Specifically as follows:

[0040] Please refer to Figure 1 As shown, the present invention provides a human-computer interaction system for an agricultural product system based on the Internet of Things, including a data acquisition module 100, a data transmission module 200, a model construction module 300, and a screen display module 400;

[0041] The data acquisition module 100 uses multiple sensors of the same type but different precisions to collect the soil humidity, temperature, and light intensity of the growth environment of agricultural products, and uses a movable high-definition camera to take pictures of the external characteristics of agricultural products, and collect the leaf color characteristic data, shape characteristic data, and leaf spot area data of agricultural products. After organizing and packing the collected data, it is sent to the data transmission module 200;

[0042] The data transmission module 200 uses a variety of wireless communication technologies for data transmission, mainly 5G mobile communication technology, supplemented by Wi-Fi and Zigbee. It adopts a hybrid backup mode and sets up a Wi-Fi hotspot locally as a backup. When the 5G signal is unstable or fails, the device can automatically switch to the Wi-Fi hotspot for data transmission. Between the device and some local short-range sensors, Zigbee can be used for communication, and the collected data is transmitted to the model construction module 300;

[0043] The model construction module 300 establishes an input layer with soil humidity, temperature, light intensity, as well as leaf color feature data, shape feature data, and leaf spot area data as input nodes, sets multiple hidden layers, each layer containing a certain number of neurons. The neurons learn the complex non-linear relationships between input features through the ReLU activation function. Through a large amount of training data, the neural network can automatically adjust the weights between the neurons in the hidden layer. The output layer sets multiple nodes to represent different categories or degrees of plant growth status, and the output results are presented in the form of probabilities. According to the output results, corresponding adjustment strategies are formulated, and the output results and the corresponding adjustment strategies are transmitted to the screen display module 400;

[0044] The screen display module 400 displays the corresponding image information according to the output results. At the same time, it uses voice broadcast to play the output results and adjustment strategies to notify the staff to make corresponding adjustments.

[0045] As Figure 2 shown, among them, the model construction module 300 includes a model creation unit 310 and an adjustment strategy unit 320;

[0046] The model creation unit 310 takes soil humidity, temperature, light intensity, as well as leaf color feature data, shape feature data, and leaf spot area data as input, and takes different categories or degrees of plant growth status as output to establish a neural network model, and divides the sorted data into a training set, a validation set, and a test set according to a certain ratio to train the neural network model;

[0047] The adjustment strategy unit 320 selects the corresponding adjustment method according to the multi-node agricultural product growth status output by the neural network model, and selects the corresponding adjustment amplitude according to the probability of the output results;

[0048] Among them, when dividing the sorted data, first perform a random shuffling operation on all the sorted data to avoid the influence of a certain order pattern that may exist in the original collection or sorting process of the data on the division result, ensuring that each data subset can evenly contain various types of data samples. Calculate the index position corresponding to 70% of the total data volume. For example, if there are a total of 1000 data entries, then the size of the training set should be 700 data entries. Through index operations, the first 700 data entries are divided into the training set, and these data will be used for the main training process of the model to enable the model to learn the relationship between input features (such as soil moisture, temperature, light, plant appearance features, etc.) and output labels (plant growth status categories).

[0049] Next, from the remaining data (i.e., data with indices 701 and later), select 20% of the total data volume as the validation set. Continuing with the above example, if there are 300 remaining data entries, then the size of the validation set should be 200 data entries. The validation set is mainly used to adjust and optimize the hyperparameters of the model (such as the number of neurons in the hidden layer, learning rate, etc.) during the training process to prevent the model from overfitting. By continuously evaluating the performance of the model on the validation set and adjusting the hyperparameters according to the changes in the performance metrics, the best model configuration can be found.

[0050] Finally, the remaining data constitutes the test set. The remaining 100 data entries are the test set. The test set is used to finally evaluate the performance of the model on unseen data after the model training is completed. It can truly reflect the generalization ability of the model, that is, the prediction accuracy of the model for new data.

[0051] In order to better predict the difference between the probability distribution and the true probability distribution, among them, when creating the model unit 310 during the training of the neural network model, the multi-class cross-entropy loss function is used to measure the difference between the model's predicted probability distribution and the true probability distribution, and the neural network model is trained with this minimum loss function as the objective. Among them, the calculation formula of the multi-class cross-entropy loss function is:

[0052]

[0053] where y p is the probability distribution output by the model, y t is the probability distribution of the true label, and C is the number of classes.

[0054] For example, a small training dataset is prepared, which contains 10 samples. Each sample has 4 input features (soil humidity, air temperature, light intensity, and plant appearance features) and a corresponding true label (plant growth status category). That is, the input features are [0.6 (soil humidity), 25 (air temperature, unit: degree Celsius), 5000 (light intensity, unit: lux), 0.8 (quantified value of plant appearance features)], and the true label is [0, 1, 0] (indicating "medium" growth status). Taking the first training sample as an example, its input features are input into the neural network model. Through the weighted sum from the input layer to the hidden layer and the ReLU activation function calculation, the output of the hidden layer is obtained. Multiply the output of the hidden layer by the weights of the output layer, and then through the softmax function calculation, the probability distribution of the output layer [0.2, 0.5, 0.3] is obtained;

[0055] For this sample, the true label is [0, 1, 0], and the predicted probability distribution is [0.2, 0.5, 0.3]. Calculate the loss:

[0056] L = -(0×log(0.2) + 1×log(0.5) + 0×log(0.3)) = -log(0.5) ≈ 0.69

[0057] After calculating the loss, it is necessary to update the weights of the neural network model according to the loss, which is achieved through the backpropagation algorithm. First, calculate the gradient of the loss with respect to the weights of the output layer. According to the calculated gradient, use the Adam optimizer to update the weights. Repeat the above steps, perform forward propagation, loss calculation, and backpropagation to update the weights for each sample in the training set. After completing one training epoch, the model has learned the entire training set once. Usually, multiple training epochs are performed until the performance of the model on the validation set (a part of the dataset divided for validating the model performance) no longer improves or reaches the preset number of training times.

[0058] In order to better adjust the learning rate and update parameters, among them, when creating the model unit 310 to train the neural network model, the Adam optimization algorithm combines the advantages of the momentum method and the adaptive learning rate, calculates the first-order moment estimate and the second-order moment estimate of the gradient, and then adjusts the learning rate and updates the parameters according to these two estimates;

[0059] The core idea of the Adam optimization algorithm is to maintain an adaptive learning rate for each parameter and use the first-order moment and second-order moment information of the gradient to accelerate convergence and reduce oscillations during training;

[0060] After a training sample (or a small batch of training samples) is input into the neural network for forward propagation and loss calculation, the gradient of the loss function with respect to the parameters is calculated through backpropagation. Assume that the currently calculated gradient is gt , update the first - order moment estimate \(m\) according to the following formula:

[0061] \(m\) t =\(\beta\) 1 \(\times m\) t-1 +(1 - \(\beta\) 1 )\(\times g\) t

[0062] where \(m\) t is the first - order moment estimate of the current step, \(m\) t-1 is the first - order moment estimate of the previous step, \(g\) t is the gradient, and \(\beta\) 1 is the weighted average.

[0063] Update the second - order moment estimate \(v\) according to the following formula:

[0064]

[0065] where \(v\) t is the second - order moment estimate of the current step, \(v\) t-1 is the second - order moment estimate of the previous step, is the square of the gradient, and \(\beta\) 2 is the weighted average.

[0066] The second - order moment estimate is used to measure the variance of the gradient and capture the change of the gradient by taking the weighted average of the square of the gradient. In the training of the neural network model for plant growth status constructed with soil humidity, air temperature, light, and plant appearance, after the input and backpropagation of each training sample (or mini - batch sample), the parameters of the model are updated using the Adam optimization algorithm according to the above steps. After multiple training cycles, the parameters of the model will gradually converge to a better value, thereby improving the prediction ability of the model for plant growth status.

[0067] In order to better adopt the corresponding priority adjustment strategy, among them, the adjustment strategy unit 320 adopts a risk - assessment - based method, combines the problem occurrence probability output by the neural network model with the harm degree, to determine the priority of different categories of growth status, and preferentially adopts the corresponding adjustment strategy;

[0068] For example, for the pest and disease problem, if the occurrence probability is 0.8 and the harm degree score is 8 points, then the risk value is 0.8×8 = 6.4; for the insufficient light problem, the occurrence probability is 0.6 and the harm degree score is 4 points, and the risk value is 0.6×4 = 2.4. Determine the priority according to the size of the risk value. The higher the risk value, the higher the priority, and the earlier the adjustment order, so as to select the priority of the corresponding adjustment strategy according to the harm degree.

[0069] Among them, the data acquisition module 100 includes a sensor acquisition unit 110 and an image acquisition unit 120;

[0070] The sensor acquisition unit 110 uses multiple sensors of the same type but different precisions to collect the soil humidity, temperature, and light intensity of the agricultural product growth environment;

[0071] The image acquisition unit 120 uses a movable high-definition camera to take pictures of the appearance features of agricultural products, and extracts the leaf color feature data, shape feature data, and leaf spot area data of agricultural products;

[0072] For example, when detecting the soil humidity of the agricultural product growth environment, a capacitive soil humidity sensor is used, and its precision is selected as high precision, medium precision, and low precision, that is, a 2% precision capacitive soil humidity sensor, a 3% precision capacitive soil humidity sensor, and a 5% precision capacitive soil humidity sensor. The capacitive soil humidity sensors with different precisions are distributed in the soil near the agricultural products to detect the soil humidity of the agricultural product growth environment, and the detected results are fused to obtain the most accurate soil humidity data.

[0073] In order to be able to determine the weights of sensors of the same type but different precisions, among them, the sensor acquisition unit 110 adopts the Kalman filter data fusion algorithm, and dynamically adjusts the weights of sensors of the same type but different precisions according to the real-time data quality and historical data of the sensors;

[0074] Kalman filtering is an optimal estimation theory used to fuse the measurement data of different sensors and perform an optimal estimation of the system state in the presence of noise. It is based on a linear system state space model, assuming that the system state transition and measurement process are linear, and the system noise and measurement noise are Gaussian distributed. Its basic idea is to continuously correct the estimation of the system state through two steps: prediction and update. The prediction step uses the dynamic model of the system to predict the state at the next moment; the update step combines the current measurement value to correct the predicted state, thereby obtaining a more accurate estimation;

[0075] A plant growth status monitoring system that uses multi-sensor data fusion of soil moisture, air temperature, light, and plant appearance continuously fuses the measurement data of each sensor according to the prediction and update steps of the Kalman filter to obtain the optimal estimate of the plant growth state. For example, in the initial stage, it may mainly rely on the predicted value. However, as the measured values are continuously input, the Kalman filter adjusts the weights of the data of each sensor according to the reliability of the measurement (determined by the Kalman gain), thereby achieving more accurate data fusion and plant growth state estimation. The Kalman filter can effectively handle the multi-sensor data fusion problem and improve the accuracy and reliability of data in fields such as plant growth monitoring, providing stronger support for subsequent decisions (such as irrigation, fertilization, light adjustment, etc.).

[0076] In order to better detect the working state of the communication link, the data transmission module 200 establishes a regular detection mechanism for the communication link. By sending test data packets and monitoring the link state, it determines whether the main communication technology is working properly. If it is not in the normal working state, the communication link is switched.

[0077] It determines whether the communication is reachable by sending an ICMP Echo Request packet to the target device and waiting for the target device to return an ICMP Echo Reply packet. The size of the Ping packet can be customized, usually defaulting to 32 bytes or 64 bytes. This kind of packet is simple and lightweight and is mainly used to detect the connectivity of the network layer. For example, sending a Ping packet from the monitoring center to the soil moisture sensor in the field. If the response returned by the sensor can be received, it indicates that the connection of the network layer is normal at least at the physical link and basic network configuration levels. A Ping packet can be sent regularly (such as every 10 minutes) to continuously monitor the communication state and ensure that the communication link is in the normal working state. Otherwise, switch to Wi-Fi communication.

[0078] As Figure 3 shown, the screen display module 400 includes an image display unit 410 and a voice broadcast unit 420;

[0079] The image display unit 410 displays the output result of the neural network and the corresponding adjustment strategy in the form of a moving picture in the middle of the screen;

[0080] The voice broadcast unit 420 broadcasts the output result of the neural network and the corresponding adjustment strategy to notify the staff to make corresponding adjustments. At the same time, it receives the voice commands of the staff and, according to the commands, selects to broadcast the key measures in the adjustment strategy;

[0081] Using the image display unit 410, the growth status of agricultural products can be intuitively reflected, as well as the adjustments that need to be made next, and it is played through the voice broadcast unit 420, which is convenient for the staff to understand. At the same time, the voice commands of the staff are recognized, reducing operations and facilitating the use of the staff.

[0082] In summary, the working principle of this solution is as follows:

[0083] This human-computer interaction system for an agricultural product system based on the Internet of Things uses the data acquisition module 100 to collect the soil humidity, temperature, and light intensity of the growth environment of agricultural products by using multiple sensors of the same type with different precisions. The Kalman filter data fusion algorithm is adopted to dynamically adjust the weights of sensors of the same type with different precisions according to the real-time data quality and historical data of the sensors. Multiple sensors jointly collect the growth environment data of agricultural products, improve the accuracy of the collected data, and reduce the influence of environmental factors on the data collected by the sensors. The data transmission module 200 uses a variety of wireless communication technologies for data transmission, mainly 5G mobile communication technology, supplemented by Wi-Fi and Zigbee, and adopts a hybrid backup mode to ensure that the data transmission module is always in a normal working state, ensure the stability of data transmission, and improve the reliability of data transmission. The model construction module 300 constructs a neural network model of the growth status of agricultural products, uses this model to judge different categories or degrees of the growth status of agricultural products, and formulates corresponding adjustment strategies according to the output results, timely adjusts the growth environment of agricultural products, ensures that agricultural products can grow normally, and increases the income.

[0084] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and do not limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A human-computer interaction system for agricultural products based on the Internet of Things, characterized by: It comprises a data acquisition module (100), a data transmission module (200), a model building module (300) and a screen display module (400); The data acquisition module (100) uses a plurality of sensors of the same type but with different precisions to collect soil moisture, temperature and light intensity of the growing environment of the agricultural products, and uses a movable high-definition camera to take pictures of the appearance characteristics of the agricultural products, collects leaf color characteristic data, shape characteristic data and leaf spot area data of the agricultural products, and organizes and packages the collected data before sending it to the data transmission module (200); The data transmission module (200) uses a variety of wireless communication technologies for data transmission, mainly using 5G mobile communication technology, supplemented by Wi-Fi and Zigbee, and adopts a hybrid backup mode, setting a Wi-Fi hotspot locally as a backup. When the 5G signal is unstable, the device can automatically switch to the Wi-Fi hotspot for data transmission. Zigbee can be used for communication between the device and some local short-range sensors, and the collected data is transmitted to the model construction module (300); The model building module (300) uses soil moisture, temperature and light intensity as well as leaf color feature data, shape feature data and leaf spot area data as input nodes to establish an input layer, and sets a plurality of hidden layers, each layer comprising neurons, and neurons learn the complex nonlinear relationship between input features through the ReLU activation function. Through a large amount of training data, the neural network can automatically adjust the weights between neurons in the hidden layer. The output layer sets a plurality of nodes to represent different categories of plant growth conditions, and the output results are presented in the form of probability. According to the output results, a corresponding adjustment strategy is formulated, and the output results and the corresponding adjustment strategy are transmitted to the screen display module (400); The screen display module (400) displays corresponding image information according to the output result, and at the same time, uses voice broadcast to play the output result and adjustment strategy to notify the staff to make corresponding adjustments.

2. The human-computer interaction system for agricultural products based on the Internet of Things according to claim 1 is characterized by: The model building module (300) includes a model creation unit (310) and a strategy adjustment unit (320); The model creation unit (310) uses soil moisture, temperature and light intensity as well as leaf color feature data, shape feature data and leaf spot area data as inputs, and uses different categories or degrees of plant growth conditions as outputs to establish a neural network model, and divides the sorted data into a training set, a validation set and a test set in proportion to train the neural network model; The adjustment strategy unit (320) selects a corresponding adjustment method according to the growth status of the multi-node agricultural products output by the neural network model, and selects a corresponding adjustment range according to the probability of the output result.

3. The human-computer interaction system for agricultural products based on the Internet of Things according to claim 2 is characterized by: When training the neural network model, the model creation unit (310) uses a multi-classification cross entropy loss function to measure the difference between the model prediction probability distribution and the true probability distribution, and trains the neural network model with the minimum loss function as the goal, wherein the calculation formula of the multi-classification cross entropy loss function is: Among them, y p is the probability distribution of the model output, y t is the probability distribution of the true label, and C is the number of categories.

4. The human-computer interaction system for agricultural products based on the Internet of Things according to claim 2 is characterized by: When training the neural network model, the model creation unit (310) uses the Adam optimization algorithm to combine the advantages of the momentum method and the adaptive learning rate to calculate the first-order moment estimate and the second-order moment estimate of the gradient, and then adjusts the learning rate and updates the parameters based on the two estimates.

5. The human-computer interaction system for agricultural products based on the Internet of Things according to claim 2 is characterized in that: The adjustment strategy unit (320) uses a risk assessment-based method to combine the probability of occurrence of the problem output by the neural network model with the degree of harm to determine the priorities of different categories of growth conditions and preferentially adopt corresponding adjustment strategies.

6. The human-computer interaction system for agricultural products based on the Internet of Things according to claim 1 is characterized by: The data acquisition module (100) comprises a sensor acquisition unit (110) and an image acquisition unit (120); The sensor collection unit (110) uses a plurality of sensors of the same type but with different precisions to collect soil moisture, temperature and light intensity of the agricultural product growth environment; The image acquisition unit (120) uses a movable high-definition camera to take photos of the appearance features of the agricultural product, and extracts leaf color feature data, shape feature data and leaf spot area data of the agricultural product.

7. The human-computer interaction system for agricultural products based on the Internet of Things according to claim 6 is characterized by: The sensor acquisition unit (110) adopts a Kalman filter data fusion algorithm to dynamically adjust the weights of sensors of the same type but different precisions according to the real-time data quality and historical data of the sensors.

8. The human-computer interaction system for agricultural products based on the Internet of Things according to claim 6 is characterized by: The image acquisition unit (120) extracts leaf color feature data by calculating the mean and variance of the leaf region color channel, extracts the edge contour of the leaf using an edge detection algorithm, and then extracts shape feature data, and extracts leaf spot area data using a threshold segmentation method.

9. The human-computer interaction system for agricultural products based on the Internet of Things according to claim 1 is characterized by: The data transmission module (200) establishes a regular detection mechanism for the communication link, and determines whether the main communication technology is working normally by sending test data packets and monitoring the link status. If it is not in a normal working state, the communication link is switched.

10. The human-computer interaction system for agricultural products based on the Internet of Things according to claim 1 is characterized by: The screen display module (400) comprises an image display unit (410) and a voice broadcast unit (420); The image display unit (410) displays the output result of the neural network and the corresponding adjustment strategy in the form of a dynamic image, which is displayed in the middle of the screen; The voice broadcast unit (420) plays the output results of the neural network and the corresponding adjustment strategy to notify the staff to make corresponding adjustments. At the same time, it receives the voice commands of the staff and selects and plays the key measures in the adjustment strategy according to the commands.

Images