A digging shovel for real-time detection of soil organic matter content

Abstract

The present invention provides an excavating shovel for real-time detection of soil organic matter content, comprising a detachable excavating shovel with an excavating blade at the front end and a sensor integrated cavity at the rear end; a data acquisition and communication module integrated inside the excavating shovel, comprising a near-infrared spectrum sensor that uses a V-shaped optical fiber diffuse reflection structure to collect soil reflectivity; a high-resolution camera for collecting soil surface texture images; a GPS module for recording latitude, longitude and elevation data of sampling points; a main control unit that serves as a core controller to coordinate data acquisition, processing and communication; and a data processing unit that realizes accurate soil organic matter content data based on the fusion of spectral and image information, helping workers to grasp soil conditions in real time and improve excavation production efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural machinery, and provides an excavating shovel capable of detecting soil organic matter content in real time. Background Art

[0002] Soil organic matter content is a core indicator of soil fertility and directly impacts crop growth efficiency. Traditional testing methods (such as high-temperature combustion and wet acid hydrolysis) rely on laboratory chemical analysis (referring to national standards such as GB 9834-1988 and NY / T 1121.6-2006). These methods have the following significant drawbacks: First, samples require pretreatment, chemical decomposition, and instrumental testing, with a single test cycle taking hours to days, making them inadequate for real-time field operations. Second, specialized technicians are required to operate precision instruments (such as elemental analyzers), resulting in high equipment maintenance costs and unsuitable for non-laboratory environments. Third, laboratory testing only supports discrete point sampling, making it difficult to generate continuous soil organic matter distribution maps. This results in a lack of global data support for fertilization decisions.

[0003] In recent years, near-infrared spectroscopy (NIRS) technology and image recognition technology have been introduced into the field of soil testing, but the following technical bottlenecks still exist: 1. Low equipment integration: Existing portable spectrometers only support single spectral data acquisition, and are not integrated with high-resolution image sensors, GPS modules, and agricultural machinery adaptation structures, resulting in low field operation efficiency. For example, operators need to manually collect samples and test them separately after agricultural machinery operations, and cannot achieve an integrated process of "analyzing while working"; 2. Single data model: Existing technologies mostly use linear regression models (such as partial least squares regression PLSR) to process spectral data, and do not integrate multi-source information such as soil surface texture (such as grayscale co-occurrence matrix features) and color distribution (RGB histogram statistics). The prediction error rate is generally ≥8%, which is difficult to meet the needs of precision agriculture; 3. Insufficient real-time performance: The end-to-end response time of solutions that rely on cloud servers for data processing exceeds 10 seconds, making it impossible to support real-time operational decisions for agricultural machinery (such as dynamically adjusting fertilizer application rates). Fourth, poor environmental adaptability: High dust and humidity in the field easily lead to sensor contamination. Existing equipment protection designs (such as bag filters) require frequent cleaning (every two hours) and are not optimized for operating conditions such as vibration and impact, resulting in insufficient equipment reliability. Furthermore, existing integrated sensors for agricultural machinery (such as soil moisture and temperature sensors) are mostly standalone modules, lacking a multimodal data fusion architecture, and have inefficient dust control designs (traditional atomization dust removal solutions have a dust rejection rate of ≤92%). These issues severely restrict the large-scale application of soil organic matter testing technology in agricultural mechanization.

[0004] In summary, existing technologies have not yet solved core problems such as low real-time field detection accuracy, weak multi-source data fusion capabilities, and poor environmental adaptability of agricultural machinery embedded systems. There is an urgent need for an innovative solution that combines laboratory-level accuracy with field operation efficiency. Summary of the Invention

[0005] The purpose of the present invention is to provide an excavating shovel for real-time detection of soil organic matter content, which can realize accurate soil organic matter content data based on the fusion of spectral and image information, grasp the soil condition in real time during actual operation, and then make more scientific and reasonable fertilization decisions, thereby improving production efficiency, avoiding resource waste, and achieving more accurate soil nutrient management.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A digging shovel for real-time detection of soil organic matter content, comprising:

[0008] The excavating shovel has an excavating blade at the front and a sensor integrated cavity at the rear, and is connected to the agricultural implement through a fixing bolt and a connecting rod;

[0009] A dustproof assembly, including a dust cover and an air pressure balancing valve built into the dust cover;

[0010] The main control unit coordinates data acquisition, communication and processing;

[0011] Global Positioning System (GPS) module, recording the latitude, longitude and elevation data of the sampling points;

[0012] A data acquisition and communication unit is integrated into the sensor integrated cavity and includes: a near-infrared spectral sensor for collecting soil reflectance, covering a wavelength range of 400-2500 nm; a camera for collecting soil surface texture images, equipped with a light source and a retractable sunshade;

[0013] Data processing unit, including:

[0014] Spectral data processing module: The envelope removal method is used to eliminate spectral baseline drift, and then the three-dimensional correlation coefficient method is used to screen sensitive bands that are strongly correlated with organic matter. The correlation between soil reflectance and organic matter content is between 0.6 and 1, which is considered a strong correlation.

[0015] Image data processing module: extracts the contrast, energy, entropy and color histogram statistics of the gray-level co-occurrence matrix (GLCM);

[0016] Multi-source data fusion model: A three-branch convolutional neural network (3D CNN) is used to fuse spectral, texture, and color features to output predicted values ​​of soil organic matter content;

[0017] Result output module: Generates soil organic matter distribution map and displays it in real time on the display screen, and uploads it to the management platform simultaneously.

[0018] As a more optimal technical solution of the present invention, the sensitive wavelength band strongly correlated with organic matter includes a combination of 918 nm, 580.8 nm and 831.8 nm.

[0019] As a more optimal technical solution of the present invention, the color histogram statistics include the mean, standard deviation, skewness and kurtosis of the three RGB channels.

[0020] As a more optimal technical solution of the present invention, the three-branch convolutional neural network includes:

[0021] a Spectral feature branch, where the spectral data filtered by the three-dimensional correlation coefficient method is processed by a 1D convolution layer;

[0022] b Texture feature branch, the texture feature parameters of the soil image extracted by the gray-level co-occurrence matrix processed by the 1D convolution layer;

[0023] c. Color feature branch, where the color features of the soil image extracted by the color histogram are processed by a 1D convolutional layer;

[0024] The above features are subjected to trilinear pooling and multi-source data fusion to establish a 3D CNN model.

[0025] As a more optimal technical solution of the present invention, the soil organic matter distribution map generated by the result output module satisfies:

[0026] Spatial interpolation: Generate 10m×10m raster data based on Kriging interpolation method;

[0027] Visual mapping: Using the HSL (hue, saturation, lightness) color space, the purple to red gradient represents 0% to 5% organic matter content.

[0028] As a more optimal technical solution of the present invention, the main control unit is a Raspberry Pi, equipped with a TensorRT inference engine, with an end-to-end response time of ≤2 seconds, and manages the continuous spectral data stream through a circular buffer.

[0029] As a more optimal technical solution of the present invention, the adjustment range of the adjustable angle connecting rod is 0°-45°, which is suitable for the interface of rotary tiller and seed drill, and the carbon steel material of the fixing bolt has a load-bearing capacity of ≥50 kg.

[0030] As a more preferred technical solution of the present invention, the near-infrared spectrum sensor, the camera and the sensor integrated cavity are detachably connected, including:

[0031] (a) The near-infrared spectral sensor is connected to the sensor integrated cavity through a base with a threaded interface;

[0032] (b) The camera is connected to the sensor integrated cavity through a slide rail.

[0033] As a more optimal technical solution of the present invention, the power supply module is a replaceable lithium-ion battery pack or a solar thin-film battery, and its battery management system BMS has the following features:

[0034] aDynamic power management, giving priority to powering the spectrum sensor and GPS module;

[0035] b. Predict the remaining power based on the operation trajectory of agricultural machinery and provide an early warning 30 minutes in advance.

[0036] As a more optimal technical solution of the present invention, the annular dust cover has a built-in air pressure balancing valve, which automatically opens the balancing airflow when the pressure difference with the external environment is ≤10 Pa.

[0037] The beneficial effects are as follows:

[0038] This invention, through the synergistic breakthrough of "multimodal data fusion architecture (spectrum + image + GPS)" and "embedded real-time detection system for agricultural machinery", introduces the three-dimensional dynamic weight distribution of spectrum, image texture and spatial position into the field of soil detection for the first time; through the detachable structure and modular design, it achieves the unity of laboratory-level precision and field operation efficiency; the combination of a three-branch CNN model and the TensorRT acceleration engine realizes algorithm-hardware collaborative optimization, defining a new standard for real-time detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a structural diagram of an excavating shovel.

[0040] Figure 2 This is a diagram of the spectral module structure inside the excavator shovel.

[0041] Figure 3 This is an example of three-dimensional correlation coefficient screening.

[0042] Figure 4 It is the result predicted by the soil organic matter prediction model, and can be used to draw a distribution image of the soil organic matter content later.

[0043] Among them: 1. Display screen; 2. Power supply module; 3. Near-infrared spectrum sensor; 4. Sunshade; 5. GPS module; 6. Connecting rod; 7. Main control unit; 8. Transmission line; 9. Data acquisition and communication module; 10. Camera; 11. Excavator shovel; 12. Fixing bolt; 13. Dust cover. DETAILED DESCRIPTION

[0044] The present invention is further described in detail below with reference to specific embodiments.

[0045] The present invention designs a deep loosening and digging shovel that integrates a spectral sensor, a camera, a control unit, and a GPS positioning module. In actual operation, the spectral data and image information of the soil can be collected while digging the soil, and the soil organic matter content can be inverted in real time through a data fusion algorithm. In addition, the digging shovel provided by the present invention can also record the geographical location of each sampling point while collecting data, generate a soil organic matter content distribution map, and provide agricultural managers with accurate soil quality data to assist in agricultural decisions such as fertilization, irrigation, and soil improvement. It solves the problems of poor timeliness, cumbersome operation, and bulky equipment in existing soil organic matter monitoring technologies, especially in the agricultural production process, and can quickly and accurately obtain soil organic matter content data and apply it efficiently in field environments.

[0046] Example 1: Excavating shovel structure and hardware configuration.

[0047] like Figure 1 As shown, the excavating shovel for real-time soil organic matter detection of the present invention comprises the following components: a detachable excavating shovel 11: with a tungsten carbide alloy excavating blade at the front and a sensor integrated cavity at the rear, measuring 120 mm × 80 mm × 50 mm (length × width × height). It features an IP67 sealing design and a built-in shock-absorbing bracket with a spring stiffness coefficient of 5 N / mm. It connects to agricultural machinery via a quick-release fixing bolt 12 (M8 bolt, pre-tightening torque 15-20 N·m) and an adjustable connecting rod 6 (adjustable range 0°-45°, locking force ≥ 200 N). A data acquisition and communication module 9: a near-infrared spectral sensor 3: employing a V-shaped fiber optic diffuse reflectance structure, covering the wavelength range of 400-2500 nm, using a 10W halogen lamp as the light source, and calibrated every 30 minutes using a Labsphere Spectralon standard whiteboard. The optical fibers are arranged with a core diameter of 400 μm (light source fiber) and 600 μm (detection fiber), forming a 60° angle, effectively collecting data from a soil depth of 2-5 mm.

[0048] High-resolution camera 10: equipped with a 2-megapixel CMOS sensor, a ring-shaped LED light source, and a retractable sunshade 4, supports ±5 cm axial slide adjustment, and collects soil surface texture images (resolution 1920×1080).

[0049] GPS module 5: uses the Ublox NEO-M8N module, with a positioning accuracy of ≤ 2 cm, and records the latitude, longitude, and elevation data of the sampling points in real time.

[0050] The main control unit 7 is implemented based on Raspberry Pi and is equipped with a TensorRT inference engine. The end-to-end response time is ≤2 seconds. The continuous data stream is managed through a circular buffer. The data acquisition and communication module 9 transmits data to the main control unit 7 through the transmission line 8.

[0051] Dust-proof components: including an annular dust cover 13 with a hydrophobic nano-coating and an air pressure balance valve (nylon 66 valve body, silicone diaphragm thickness 0.5 mm), which automatically opens when the internal and external air pressure difference is ≥8 Pa, and the dust blocking rate is ≥99.2%.

[0052] Example 2: Data processing flow.

[0053] Spectral data preprocessing includes:

[0054] Reflectivity calculation: by formula CR ( l ) = ( I ( l )-min( I )) / (max( I )- min( I )) Eliminate ambient light interference, where I ( l ) is the spectral signal at wavelength l The value below, min( I ) and max( I ) are the minimum and maximum values ​​of the signal, CR ( l ) is the wavelength l The envelope value under ;

[0055] Sensitive band screening: Based on the Pearson correlation coefficient (r>0.85), 918 nm (C-H bond stretching vibration), 580.8 nm (N-H bond deformation), and 831.8 nm (O-H bond bending) were selected as characteristic bands.

[0056] Image feature extraction:

[0057] Texture features: GLCM is used to calculate contrast, energy, entropy, homogeneity, correlation and angular second moment;

[0058] Color features: Extract the mean, standard deviation, skewness, and kurtosis of the RGB channels.

[0059] Example 3: System performance verification.

[0060] Experimental conditions: Mounted on a tractor in the black soil region of Northeast China (organic matter content 2.1%-4.8%), operating speed 5 km / h, 1200 sets of samples were collected continuously.

[0061] Result analysis: Prediction accuracy: R 2 =0.94, RMSE=0.31g / kg (42% reduction in error compared to traditional methods).

[0062] Real-time performance: End-to-end response time is 1.6 seconds (95% confidence interval), supporting processing of 100 data points per second.

[0063] Anti-interference: Under 5-8G vibration conditions, the prediction error fluctuation rate is ≤3% (traditional model ≥12%).

[0064] Example 4: Generation of soil organic matter distribution map.

[0065] Spatial interpolation: Kriging interpolation (spherical model, range 35 m) was used to generate 10 m × 10 m grid data.

[0066] Visual Mapping: Display organic matter content gradients in HSL color space (purple to red corresponds to 0%-5%) and export reports in PDF or CSV format.

[0067] Example 5: Modular expansion application.

[0068] Power supply module 2: supports replaceable lithium-ion battery packs (8 hours of battery life) or solar thin-film batteries. The BMS system predicts power consumption based on the trajectory of the agricultural machinery and provides a 30-minute advance warning.

[0069] Algorithm adaptation: For red soil areas, the sensor acquisition frequency is dynamically adjusted to 2 times (when soil moisture > 60%), and model parameters are optimized through transfer learning.

[0070] Example 6: Dustproof and durability test.

[0071] After 8 hours of continuous operation, the dust content in the sensor cavity was ≤0.1 mg / m³, and the contact angle on the dust cover was ≥150°, indicating that the hydrophobic performance had not degraded. The polyurethane damping pad (Shore A hardness of 70A) on the connecting rod effectively absorbed 83% of high-frequency vibrations, extending the component life to 2000 hours.

[0072] The above embodiments show that the present invention achieves efficient and accurate detection of soil organic matter content through multi-sensor integration, advanced algorithm fusion and modular design, providing reliable technical support for precision agriculture.

[0073] By combining a near-infrared spectral sensor with an image acquisition system, this method can acquire spectral and image information of the soil in real time, enabling immediate monitoring of soil organic matter content. This method avoids the time delays and data feedback lags associated with traditional methods, helping agricultural workers obtain timely soil information and quickly implement appropriate agricultural measures.

[0074] By integrating a spectral sensor, image acquisition device, and control unit into a digging shovel that can be connected to agricultural machinery, this invention allows agricultural workers to conduct soil testing directly in the fields without having to carry bulky laboratory equipment, significantly improving agricultural production efficiency.

[0075] This method uses a multi-band spectral sensor and a high-resolution camera, combined with advanced data processing algorithms (three-dimensional correlation coefficient screening and a three-branch convolutional neural network that fuses multi-source information) to achieve high-precision prediction of soil organic matter content. Through data fusion technology, spectral data and image information complement each other, providing a more comprehensive picture of the soil's true condition and reducing measurement errors.

[0076] The system, combined with a GPS positioning module, can record the geographic location of each sampling point in real time and generate soil organic matter distribution maps. These charts not only provide agricultural managers with intuitive soil quality information, but can also be used to monitor and analyze soil quality trends, supporting precise fertilization and soil improvement decisions.

[0077] The excavating shovel design of the present invention is highly modular and customizable, and can be configured according to different soil types, crop requirements and agricultural environments. It is suitable for various agricultural operation scenarios, especially in the field of precision agriculture, and has broad application prospects.

[0078] The excavating shovel of the present invention can be mechanically connected to an operating agricultural machine. The excavating shovel head excavates the soil to obtain samples, while the spectral sensor and camera simultaneously collect spectral information and surface images of the soil. A control unit transmits this data to a processing unit, which uses an algorithm to analyze the spectral and image features to extract predicted data on organic matter content. A GPS module records the sampling location, and the system combines the soil data and location information to generate a soil organic matter distribution map. Finally, the control unit stores this data and exports it as a report or distribution map.

[0079] This invention combines spectral analysis, image processing, data fusion, and GPS positioning technologies to provide a portable soil organic matter detection tool that is convenient for agricultural workers. By optimizing the design of a deep-soiling shovel, the tool collects spectral data and image information from the soil in real time while digging. Using a built-in control unit and advanced algorithms, it accurately predicts the soil organic matter content.

[0080] The present invention provides an excavating shovel for real-time detection of soil organic matter content, comprising: a detachable excavating shovel 11, with an excavating blade at the front end and a sensor integrated cavity at the rear end, connected to the agricultural implement via a quick-release fixing bolt 12 and an angle-adjustable connecting rod 6; a data acquisition and communication module integrated into the excavating shovel, comprising: a near-infrared spectral sensor 3, which uses a V-shaped optical fiber diffuse reflection structure to collect soil reflectivity, covering a wavelength range of 400-2500 nm; a high-resolution camera 10, equipped with a ring-shaped LED light source and a retractable sunshade 4, for collecting soil surface texture images; a GPS module 5, which records the latitude, longitude and elevation data of the sampling point, with a positioning accuracy of ≤2 cm; a Raspberry Pi. P serves as the core controller to coordinate data acquisition, processing and communication; the data processing unit includes: a spectral data processing module: which uses the envelope removal method and three-dimensional correlation coefficient screening to preprocess and extract features of the spectral data; an image data processing module: which extracts the contrast, energy, entropy and color histogram statistics of the gray-level co-occurrence matrix; a multi-source data fusion model: which uses a three-branch convolutional neural network to fuse spectral, texture and color features and outputs a predicted value of soil organic matter content; a result output module: which generates a soil organic matter distribution map and displays it in real time on the display screen, and simultaneously uploads it to the management platform; a dustproof component, including a hydrophobic nano-coated annular dust cover 13 and an air pressure balance valve, which reduce field dust pollution.

[0081] The excavating shovel design of the present invention is based on a detachable deep-soiling excavating shovel, combined with a spectral sensor, an image camera, and a control unit, and has the following features:

[0082] The excavator head is a standard excavator head with multiple sensors integrated into its rear, including a near-infrared spectral sensor and a visible light image camera. The excavator head's design ensures simultaneous acquisition of soil spectral data and surface image information during excavation. The spectral module consists of a light source, a fiber optic transmission system, and a photoelectric sensor. The module includes multiple V-shaped optical fibers that transmit light. The light is diffusely reflected by the soil and then reaches the photoelectric sensor, which converts the optical signal into an electrical signal. The signal processing unit performs amplification, filtering, and A / D conversion to ensure high-quality and low-noise data. This module can also provide spectral information for inverting soil organic matter. The camera is also equipped with a high-resolution visible light camera to capture images of the soil surface. The camera uses image processing algorithms to extract soil surface features, using a gray-level co-occurrence matrix to extract soil surface texture features, and a color histogram to extract soil image color features, helping to improve the accuracy of soil organic matter content prediction. The control unit is the core component of the excavator head, responsible for collecting, processing, and calculating sensor data. Specifically, it includes the following parts: Hardware part: The control unit uses Raspberry Pi as the main control unit to process signals collected from spectral sensors, cameras and other devices. Raspberry Pi has strong computing power, can process large amounts of data in real time, and effectively connect with other modules. Software part: The software part uses Python programming language to realize data acquisition and processing, image processing, data fusion and soil organic matter prediction. This part includes the following functions: Data acquisition: Soil spectral data and image data are collected from spectral sensors and cameras, and the collected data is transmitted to the control unit. Spectral data processing: First, the spectral data is pre-processed using the de-envelope method, and the high correlation bands are selected using the three-dimensional correlation coefficient screening method; some index calculation formulas and screening results are as follows:

[0083] CR ( l ) = ( I ( l )-min( I )) / (max( I )- min( I ))

[0084] in I ( l ) is the spectral signal at wavelength l The value below, min( I ) and max( I ) are the minimum and maximum values ​​of the signal, CR ( l ) is the wavelength l The envelope value below.

[0085] .

[0086] R1, R2, and R3 represent reflectance data from different spectral bands. These bands were selected based on correlation analysis of soil organic carbon (SOC) content. The purpose of selecting these bands was to identify the wavelength combination with the strongest correlation with SOC content. After the data were transformed using TDI, the correlation coefficient between the data series and the soil organic matter content series was calculated using the Pearson correlation coefficient method.

[0087]

[0088] r is the Pearson correlation coefficient. x i and y i are the observed values ​​of the two variables. and They are x and y The average value of . n is the number of samples.

[0089] Figure 3 The three-dimensional coefficient correlation diagram obtained after screening by different TDI formulas. In each diagram, three sensitive bands can be seen, such as Figure 3 The three sensitive bands of (a) are 918 mm, 580.8 nm, and 831.8 nm. After each acquisition of spectral data, the algorithm model in the control unit will calculate multiple TDIs and select the TDI model with the highest correlation as one of the input parts of the subsequent multi-source data model.

[0090] Image data processing: Soil surface features were extracted using image recognition algorithms. Gray-level co-occurrence matrix methods were used to extract soil surface texture characteristic indices (contrast, energy, entropy, homogeneity, correlation, and angular second moment). Color histogram methods were used to extract soil image color characteristic indices (red mean, green mean, blue mean, red standard deviation, green standard deviation, blue standard deviation, red skewness, green skewness, blue skewness, red kurtosis, green kurtosis, and blue kurtosis).

[0091] Three-branch convolutional neural network model for multi-source data fusion: Spectral data and image features are fused with multi-source information (spectral features, texture features, color features), and the fused data are substituted into the support vector machine model (SVM), random forest model (RF), partial least squares model (PLS), and 3D CNN model. The model results are as follows: Figure 4 shown.

[0092]

[0093]

[0094] RMSE represents the absolute error of prediction, and R² represents the correlation coefficient. In regression analysis, the closer R² is to 1, the lower the RMSE value is, which usually indicates that the model has strong fitting ability, small prediction error, and good model validity and reliability.

[0095] Soil prediction: Based on the prediction results of the soil organic matter prediction model and the recorded geographic coordinates, a distribution image of the soil organic matter content is drawn.

[0096] To ensure accurate correlation between soil data and geographic location, the present invention incorporates an integrated GPS module to record the geographic location of each sampling point in real time. Assuming a GPS module with an accuracy of ±5 meters, by recording location data in real time, the geographic coordinates of each sampling point can be precisely correlated with the collected soil data, thereby improving the accuracy of soil organic matter content distribution maps. For example, if 100 soil samples are collected across 10 hectares of farmland, precise GPS positioning allows agricultural managers to track soil quality distribution in real time, enabling precise fertilization and soil management. Combined with high-precision GPS positioning (such as the Ublox NEO-M8N module, with an accuracy of ±2.5 meters), this technology provides efficient spatial data support.

[0097] All collected soil data (including spectral data, image data, GPS positioning data, etc.) is stored in the control unit and displayed in real time on Display 1. Assuming the system collects 100 data points per second, after one hour of soil testing, a total of 360,000 data points will be collected. This data will be stored and processed in real time. Agricultural workers can view the predicted soil organic matter content in real time through the control unit, allowing them to make more precise adjustments based on the data. The system also supports exporting data in CSV or PDF format, facilitating the generation of soil organic matter distribution maps and forecast reports, further assisting agricultural management decisions.

[0098] The system of this invention offers excellent user interactivity, allowing users to intuitively view real-time collected data and predicted soil organic matter content on the display screen. The system features a simple and intuitive user interface, allowing users to select different soil samples via the touchscreen to view or modify collection parameters. Assuming the system can display and update data within one second, if the predicted soil organic matter content deviates from a predetermined threshold (e.g., by more than ±10%), the system will alert agricultural workers in real time to adjust their fertilization schedules. This intuitive user interaction allows agricultural workers to respond more quickly, improving the efficiency and accuracy of agricultural production.

[0099] The excavator shovel design of the present invention is highly modular and customizable, allowing users to select appropriate sensor and algorithm configurations based on soil type, crop requirements, and the needs of specific agricultural environments. For example, the system can dynamically adjust the sensor sensitivity and data acquisition frequency based on soil pH, humidity, temperature, and other information. When soil moisture exceeds 60%, the acquisition frequency of the spectral sensor can be doubled to ensure data accuracy. This modular design allows the device to adapt to different agricultural operation scenarios, especially in the fields of precision agriculture (such as crop yield prediction), smart agriculture (such as automated irrigation), and environmental monitoring (such as contaminated soil detection), and has a wide range of applicability. By adopting standardized interfaces, the device can expand new sensor modules or algorithms according to different needs, further improving the flexibility and adaptability of the system.

[0100] In some embodiments, the texture features extracted by the image data processing module include contrast, energy, entropy, homogeneity, correlation and angular second moment of the gray-level co-occurrence matrix; the color features include the mean, standard deviation, skewness and kurtosis of the RGB three channels.

[0101] In some embodiments, the 3D CNN includes: a spectral feature branch, which processes the spectral data filtered by the three-dimensional correlation coefficient method through a 1D convolution layer; b texture feature branch, which processes the texture feature parameters of the soil image extracted by the gray-level co-occurrence matrix through a 1D convolution layer; c color feature branch, which processes the color features of the soil image extracted by the color histogram through a 1D convolution layer. The above features are subjected to trilinear pooling and multi-source data fusion to establish a 3D CNN model. The model evaluation satisfies the determination coefficient R 2 ≥0.92 and root mean square error RMSE≤0.35g / kg.

[0102] Sensitive band verification: The Pearson correlation coefficient matrix of 918 nm (C-H bond stretching vibration), 580.8 nm (N-H bond deformation), and 831.8 nm (O-H bond bending) is shown in Table 1.

[0103] Table 1

[0104]

[0105] Comparative test with NY / T 1121.6-2006 method (n=500).

[0106] Achieving 4.3x speedup with TensorRT’s INT8 quantization: 200 representative spectral image pairs.

[0107] Quantification error: ≤0.02g / kg (based on RMSE ≤0.35g / kg).

[0108] Compared with the traditional two-branch model (such as 1D+2D CNN), this structure achieves dynamic weight distribution through a multi-head attention mechanism. The comparison on the ImageNet test set is shown in Table 2.

[0109] Table 2

[0110]

[0111] A channel attention mechanism (SE module) is used to enhance vibration noise robustness. Under 5-8G acceleration conditions, the prediction error fluctuation rate is ≤3% (traditional models ≥12%). Dynamic weight allocation of feature channels is achieved through the following three-stage operation.

[0112] Squeeze stage (global information compression):

[0113] Input feature map dimension U : (H=224, W=224, C=512 corresponds to the final residual block output of ResNet-18).

[0114] Global average pooling operation: Generate channel statistics vector ,in z c Indicates the c Global spatial characteristics of the channels.

[0115] Excitation stage (channel correlation modeling):

[0116] Dual fully connected layer structure:

[0117] Dimensionality reduction fully connected layer ( r =16 is the compression ratio, which has been verified by experiments to be the optimal balance point).

[0118] : ReLU activation function (to prevent gradient disappearance):

[0119] : Dimension-increasing fully connected layer (restore the original number of channels);

[0120] : Sigmoid function (output weight value range [0,1]);

[0121] Reweight stage (feature channel calibration):

[0122] Channel-wise multiplication: in s c Indicates the cThe importance weights of each channel are used to achieve noise suppression and key feature enhancement.

[0123] The compression ratio r is determined as shown in Table 3.

[0124] Table 3

[0125]

[0126] When r=16 is selected, the model reaches Pareto optimality in terms of parameter quantity, accuracy and speed.

[0127] Vibration noise suppression: by channel weighting.

[0128] Dynamically reduce the weight of high-frequency vibration noise-related channels (experiments show that the noise channel weight attenuation under vibration conditions is ≥70%).

[0129] Feature robustness enhancement: A 1.2-1.5 times weight gain is assigned to channels related to soil surface texture (GLCM features) and color distribution (RGB histogram).

[0130] Converting the fully connected layer weights of the SE module from FP32 to INT8 achieves a 4.3x speedup on a Raspberry Pi 4B (latency reduced from 85ms to 20ms).

[0131] A ring buffer is used to manage SE weight vectors, reducing memory usage by 62% (from 8.2MB to 3.1MB).

[0132] The effectiveness verification of channel attention is shown in Table 4.

[0133] Table 4

[0134]

[0135] The computational overhead analysis is shown in Table 5.

[0136] Table 5

[0137]

[0138] Circular buffer management supports up to 30 seconds of breakpoint resuming, adapting to intermittent loss of agricultural machinery GPS signals (average packet loss rate ≤ 2%).

[0139] In some embodiments, the soil organic matter distribution map generated by the result output module satisfies:

[0140] Spatial interpolation: Generate 10m×10m raster data based on Kriging interpolation method;

[0141] Visualization mapping: Using HSL color space, the purple to red gradient represents organic matter content from 0% to 5%.

[0142] In some embodiments, the main control unit of the data processing unit is a Raspberry Pi, equipped with a TensorRT inference engine, with an end-to-end response time of ≤2 seconds, and a continuous spectral data stream is managed through a circular buffer.

[0143] In some embodiments, the adjustable angle connecting rod 6 has an adjustment range of 0°-45°, is adapted to the interface of a rotary tiller and a seed drill, and the carbon steel material of the fixing bolt (12) has a load-bearing capacity of ≥50 kg.

[0144] In some embodiments, the near-infrared spectral sensor 3 and the camera 10 adopt a modular quick-release design, including:

[0145] (a) Spectral sensor base, equipped with an M12 threaded interface and a waterproof sealing ring;

[0146] (b) Camera slide bracket, supporting axial position adjustment of ±5 cm.

[0147] In some embodiments, the power supply module 2 is a replaceable lithium-ion battery pack or a solar thin-film battery, and its battery management system (BMS) has the following features:

[0148] (a) Dynamic power management, prioritizing power supply to the spectral sensor and GPS module;

[0149] (b) Predict the remaining power based on the operation trajectory of agricultural machinery and provide a 30-minute advance warning.

[0150] In some embodiments, the annular dust cover 13 of the dustproof assembly has a built-in air pressure balancing valve, which automatically opens the balancing airflow when the pressure difference with the external environment is ≤10 Pa.

[0151] In some embodiments, the quick-release fixing bolt 12 uses an M8 bolt with a pre-tightening torque of 15-20 N·m, and is compatible with the ISO2320 standard agricultural machinery interface.

[0152] In some embodiments, a polyurethane damping pad (Shore hardness 70A) is added between the connecting rod and the fixing bolt, which can absorb more than 80% of high-frequency vibrations (frequency > 50 Hz) during agricultural machinery operations.

[0153] In some embodiments, the end of the adjustable angle connecting rod 6 is configured with an ISO 5675 standard quick-hook, which supports the universal interface of rotary tillers (such as John Deere 5E series) and seed drills (such as Case IH 2150), with an adjustment angle accuracy of ±1° and an adjustment locking force of ≥200 N.

[0154] In some embodiments, the sensor integrated cavity has a cavity size of 120 mm × 80 mm × 50 mm (length × width × height), adopts an IP67-level sealing design, and has an internal pre-installed shock-absorbing bracket (spring stiffness coefficient 5 N / mm).

[0155] In some embodiments, the light source optical fiber (core diameter 400 μm, NA 0.22) and the detection optical fiber (core diameter 600 μm, NA 0.24) are at an angle of 60°, with a spacing of 0.5 mm, and the effective collection depth is 2-5 mm soil layer.

[0156] In some embodiments, the halogen lamp source (wavelength range 400-2500 nm, power 10 W, color temperature 2856 K) is automatically calibrated every 30 minutes (using a Labsphere Spectralon standard white plate).

[0157] In some embodiments, the base is fixed via an M12 threaded interface (compliant with IEC 61076-2-101 standard), and the waterproof sealing ring is made of fluororubber (temperature resistant from -20°C to 120°C).

[0158] In some embodiments, the spectral signal is normalized to the reflectance: R(λ) = ((λ)−D(λ)) / (W(λ)−D(λ)).

[0159] In some embodiments, the envelope connection point selection rule is to use the absorption peak minimum (such as 1410 nm, 1910 nm) as the node and perform piecewise linear fitting.

[0160] In some embodiments, the wavelength range is (400-2500 nm), the spatial range is (average of a 1 m² area around the sampling point), and the temporal range is (sliding window of 10 consecutive samples).

[0161] In some embodiments, based on the Pearson correlation coefficient (r>0.85) screening, it was experimentally verified that the combined prediction error of 918 nm (CH bond stretching vibration), 580.8 nm (NO3⁻ absorption peak), and 831.8 nm (organic aromatic C=C characteristics) was the lowest (RMSE=0.28 g / kg).

[0162] In some embodiments, the spectral branch is: 3 layers of 1D convolution (kernel size [3, 5, 3], stride 1, number of channels [32, 64, 128]), followed by a global average pooling layer.

[0163] In some embodiments, the fusion branch includes a multi-head attention mechanism (4 heads, embedding dimension 256), and the fusion weights are dynamically allocated through learnable parameters.

[0164] In some embodiments, the model training process is as follows:

[0165] Dataset: 5,000 soil samples (organic matter content 0.5%-5.2%), 70% training set, 15% validation set, and 15% test set;

[0166] Hyperparameters: initial learning rate 0.001 (Cosine decay), batch size 32, loss function Huber Loss (δ=0.5), optimizer AdamW;

[0167] Lightweight measures: FP16 quantization is used during TensorRT deployment, and the model inference time is measured to be 1.3-1.8 seconds (Raspberry Pi 4B, ambient temperature 25°C).

[0168] In some embodiments, interpolation parameters are as follows: semivariogram: spherical model (nugget value 0.12, sill value 1.05, range 35 m), anisotropy ratio 1.5 (azimuth 30°). Search strategy: maximum radius 50 m, minimum 8 neighborhood points, interpolation error controlled by cross-validation (mean absolute error MAE ≤ 0.2 g / kg).

[0169] In some embodiments, the pressure-balancing valve body is made of nylon 66 (temperature resistant -40°C to 120°C), with a built-in silicone diaphragm (thickness 0.5 mm, elastic modulus 1.5 MPa) and a spring stiffness of 0.8 N / mm. A differential pressure sensor (Honeywell HSC series) monitors the internal and external pressures in real time and triggers the valve to open when ΔP ≥ 8 Pa (response time ≤ 0.5 seconds).

[0170] This example was tested in Northeast China's black soil region (organic matter content 2.1%-4.8%), using a Lovol M904 tractor at a speed of 5 km / h. Prediction accuracy: R² = 0.94, RMSE = 0.31 g / kg (n = 1200). Response time: 1.6 seconds (95% confidence interval). Dust prevention: No dust after 8 hours of continuous operation (dust blocking efficiency 99.2%).

[0171] In summary, the present invention aims to provide a soil organic matter detection tool that is highly real-time, convenient, and accurate. By acquiring soil data in real time, it helps agricultural producers make scientific and reasonable decisions and promotes the development of agricultural production towards intelligence and precision.

[0172] The shovel head of the excavating shovel of this invention integrates multiple sensors (such as a near-infrared spectral sensor and an image camera) while also maintaining normal excavation functionality. By optimizing the structural design of the excavating shovel head, it can simultaneously collect soil spectral data and surface image information while performing soil excavation tasks. This design allows agricultural workers to obtain soil information directly in the field without carrying bulky equipment, thereby improving the convenience and efficiency of soil testing. The key features of three-dimensional sensitive band screening and 3D CNN are: first, a three-dimensional correlation coefficient screening method is used to select bands highly correlated with soil organic matter content from multiple spectral bands, ensuring the extraction of the most predictive spectral data and improving the model's prediction accuracy. Next, a three-branch convolutional neural network is used to process the fused spectral data and image features. Spectral features, texture features, and color features are independently processed, and the results are fused to achieve a more comprehensive prediction of soil organic matter content. This method effectively combines the advantages of multi-source data, reducing errors and improving the accuracy and reliability of predictions. The system can record the geographic location of each data sampling point, combine the collected soil data with the geographic location, and generate a soil organic matter distribution map to help agricultural managers make precise fertilization, irrigation and soil improvement decisions.

[0173] The spectral and image sensors in the excavating shovel of this invention can be configured to suit different soil types and environmental conditions. The spectral sensors automatically adjust to soil moisture and other environmental factors to ensure data accuracy. Furthermore, algorithms are optimized based on different soil characteristics to provide precise organic matter predictions.

Claims

1. A digging shovel for real-time detection of soil organic matter content, characterized in that: include: An excavating shovel (11) having an excavating blade at the front end and a sensor integrated cavity at the rear end, connected to the agricultural implement via a fixing bolt (12) and a connecting rod (6); A dustproof component, comprising a dustproof cover (13) and an air pressure balancing valve built into the dustproof cover (13); Main control unit (7), coordinating data collection, communication and processing; GPS module (5), recording the latitude, longitude and elevation data of the sampling points; A data acquisition and communication unit (9) is integrated into the sensor integrated cavity and includes: a near-infrared spectral sensor (3) for collecting soil reflectance, covering a wavelength range of 400-2500 nm; a camera (10) for collecting soil surface texture images, equipped with a light source and a retractable light shield (4); Data processing unit, including: Spectral data processing module: The spectral baseline drift is eliminated using the envelope removal method. Then, the three-dimensional correlation coefficient method is used to screen sensitive bands that are strongly correlated with organic matter. A correlation between soil reflectance and organic matter content between 0.6 and 1 is considered strong. The sensitive bands strongly correlated with organic matter include a combination of 918 nm, 580.8 nm, and 831.8 nm. Image data processing module: extracts the contrast, energy, entropy and color histogram statistics of the gray-level co-occurrence matrix; the color histogram statistics include the mean, standard deviation, skewness and kurtosis of the RGB three channels; Multi-source data fusion model: A three-branch convolutional neural network is used to fuse spectral, texture, and color features to output the predicted value of soil organic matter content; The three-branch convolutional neural network includes: (a) Spectral feature branch, where the spectral data filtered by the three-dimensional correlation coefficient method is processed by a 1D convolutional layer; (b) Texture feature branch, the texture feature parameters of the soil image extracted by the gray-level co-occurrence matrix processed by the 1D convolution layer; (c) Color feature branch, where the color features of the soil image extracted by the color histogram are processed by a 1D convolutional layer; The above features are subjected to trilinear pooling and multi-source data fusion to build a 3D CNN model; Result output module: Generates a soil organic matter distribution map and displays it in real time on the display screen (1), and simultaneously uploads it to the management platform; the soil organic matter distribution map satisfies: Spatial interpolation, generating 10m×10m raster data based on Kriging interpolation method; Visualization map, using HSL color space, with the purple to red gradient representing 0% to 5% organic matter content.

2. The excavating shovel for real-time detection of soil organic matter content according to claim 1, characterized in that: The main control unit (7) is a Raspberry Pi equipped with a TensorRT inference engine, with an end-to-end response time of ≤2 seconds, and manages the continuous spectral data stream through a circular buffer.

3. The excavating shovel for real-time detection of soil organic matter content according to claim 1, characterized in that: The angle adjustment range of the connecting rod (6) is 0° to 45°, and is suitable for the interface of a rotary tiller and a seed drill.

4. The excavating shovel for real-time detection of soil organic matter content according to claim 1, characterized in that: The near-infrared spectrum sensor (3) and the camera (10) are all detachably connected to the sensor integrated cavity, including: (a) The near-infrared spectral sensor (3) is connected to the sensor integrated cavity via a base with a threaded interface; (b) The camera (10) is connected to the sensor integrated cavity via a slide rail.

5. The excavating shovel for real-time detection of soil organic matter content according to claim 1, characterized in that: It also includes a power supply module (2), which is a replaceable lithium-ion battery pack or a solar thin-film battery and is used to supply power to the display screen (1), the main control unit (7), the GPS module (5) and the data acquisition and communication unit (9).

6. The excavating shovel for real-time detection of soil organic matter content according to claim 1, characterized in that: The dust cover (13) has a built-in air pressure balancing valve, which automatically opens the balancing airflow when the pressure difference with the external environment is ≤10 Pa.

Images