A self-propelled pepper harvester operation efficiency monitoring method and device

Abstract

The application provides a self-propelled pepper harvester operation efficiency monitoring method and device, and relates to the technical field of intelligent monitoring of agricultural machinery.The application obtains the driving torque and the rotation angle phase signal of a single picking cycle of a human-imitated hand flexible elastic tooth fruit picking drum, establishes a motion torque atlas, disassembles along the phase, divides the loading and unloading stages according to the zero-crossing characteristics of the torque change rate, extracts the torque climbing slope and the falling steepness to form a feature vector, inputs a quality classifier based on non-destructive picking sample calibration, realizes the automatic classification of effective non-destructive, slightly damaged and empty picking according to the feature envelope line, counts the effective fruit picking number in a sliding time window, calculates the time window harvesting efficiency in combination with the preset single fruit quality mean value, and generates continuous efficiency time series data.The method takes the picking action mechanics response analysis as the core, realizes the online fine monitoring of the operation efficiency and the synchronous evaluation of the quality, and provides reliable quantitative basis for low-damage and high-efficiency harvesting.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for agricultural machinery, specifically to a method and device for monitoring the operating efficiency of a self-propelled chili harvester. Background Technology

[0002] Self-propelled chili harvesters are the core equipment for mechanized harvesting in large-scale chili cultivation. Their field operation efficiency directly determines the harvesting progress, equipment scheduling, and the final fruit quality qualification rate. Real-time monitoring of operation efficiency not only provides operational feedback to operators but also provides farm managers with quantitative basis for evaluating operation progress and quality. In chili harvesting scenarios, due to the thin skin of the chili peppers and the uneven strength of the stem connections, fruit damage is very likely to occur during the harvesting process. Therefore, the monitoring of harvester operation efficiency cannot be limited to the harvested weight per unit time. It is also necessary to simultaneously perceive the completion quality of each picking action to avoid a large number of minor damages or invalid actions being mixed into the efficiency statistics, resulting in inflated monitoring results and a decrease in the actual marketable yield.

[0003] In the existing technology, the main methods for monitoring the operating efficiency of harvesters rely on installing weighing sensors on the conveying or fruit collection channels, or indirectly estimating the feed amount by monitoring engine speed, walking speed and cutting width, and then estimating the instantaneous operating efficiency. Such methods take the macroscopic operating parameters of the whole machine or large components as input and take the total amount of material passing through a certain cross section per unit time as the basis for efficiency calculation. In essence, it is a cumulative statistical analysis of the harvesting results and cannot establish a correlation between efficiency data and the execution quality of a single harvesting action. However, the existing technologies have the following shortcomings: First, the existing methods cannot distinguish invalid picking actions included in the efficiency statistics, such as empty swinging of the picker or brushing the stems and leaves without separating the fruit. These actions will still be included in the operation time in the macro statistics, but no effective harvest will be produced, resulting in a deviation between the efficiency estimate and the actual harvest. Second, the existing methods do not have the ability to classify the quality of picking actions online, and cannot distinguish between non-damaged picking and slightly damaged picking in the process of efficiency monitoring, so that the efficiency output value cannot reflect the degree of skin integrity of the harvested fruit.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method and device for monitoring the operating efficiency of a self-propelled chili harvester, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for monitoring the operating efficiency of a self-propelled chili harvester, comprising the following steps: S1: Obtain the real-time drive torque signal and roller rotation phase signal of a single complete picking cycle of the human hand-like flexible toothed fruit picking roller, and establish a motion torque spectrum with a single picking action as the basic unit. S2: Deconstruct the action torque spectrum along the phase signal of the drum rotation angle, and based on the zero-crossing characteristics of the torque change rate, decompose a single picking action into the loading stage of the progressive insertion of the spring teeth and the unloading stage of the pepper fruit detaching from the stem. S3: Extract the torque ramp-up slope during the loading phase and the torque drop steepness during the unloading phase of this single picking action, and combine the two to obtain the feature vector of this single picking action; S4: Input the feature vector of this single picking action into the pre-calibrated picking quality classifier. Based on the non-destructive picking feature envelope formed by the torque rise slope and torque fall steepness of the non-destructive picking sample, classify the quality of this single picking action to obtain effective non-destructive picking, effective slightly damaged picking, or invalid empty picking. S5: Set a time window, count the number of picking actions of each quality type within the time window, combine the number of effective undamaged picking and the number of effective slightly damaged picking to obtain the effective harvest quantity, calculate the quality harvest quantity with the preset average single fruit quality, and calculate the time window harvest efficiency with the quality harvest quantity and the time window duration. S6: Generate several continuous time windows along the operation time axis with a fixed sliding step size. Repeatedly execute S2 to S5 to complete the window-by-window calculation of the harvesting efficiency. Arrange the harvesting efficiency of each time window in chronological order to obtain continuously changing harvesting efficiency time series data.

[0007] Furthermore, the real-time drive torque signal output by the torque sensor and the rotational pulse signal output by the rotational encoder are synchronously acquired at a fixed sampling frequency, and the rotational pulse signal is converted into a drum rotational phase signal; The sampling time when a specific spring tooth on the human hand-like flexible spring tooth fruit picking roller passes through a fixed reference point is used as the starting reference for the roller's rotation angle, and the sampling time when the specific spring tooth completes one complete telescopic picking cycle and passes through the fixed reference point again is used as the ending reference for the roller's rotation angle. The real-time drive torque signal and the drum rotation angle phase signal, which are synchronously collected within the phase interval of the drum rotation angle start reference and the drum rotation angle end reference, are associated and stored point by point to obtain the picking action torque map corresponding to a single picking action.

[0008] Furthermore, the rate of torque change of the real-time drive torque signal in the torque spectrum of the picking action is calculated relative to the value of the drum rotation angle phase signal. The phase point of the drum rotation angle phase signal corresponding to the change rate of torque from a positive value to a zero value is defined as the end phase point of the loading phase and the start phase point of the unloading phase. The phase point of the drum rotation angle phase signal corresponding to the change rate of torque from negative to zero is defined as the end phase point of the unloading phase. The interval between the starting reference of the drum rotation angle and the ending phase point of the loading stage is divided into the loading stage; The interval between the start phase point and the end phase point of the unloading phase is defined as the unloading phase.

[0009] Furthermore, the torque ramp-up slope during the loading phase and the torque drop steepness during the unloading phase of this single harvesting action are extracted separately, with the specific logic as follows: Using the value of the drum rotation phase signal during the loading stage as the independent variable and the value of the real-time drive torque signal as the dependent variable, the corresponding data of the two are fitted with least squares, and the slope of the fitted line is determined as the torque climb slope of the picking action. Extract the peak torque value corresponding to the peak point of the real-time drive torque signal during the unloading phase, the roller rotation phase point corresponding to the peak torque value, and the roller rotation phase point corresponding to the torque value falling back to the no-load baseline. The difference between the peak torque and the unloaded baseline is used as the torque drop amplitude. The angle difference between the peak torque corresponding to the roller rotation phase point and the torque drop to the unloaded baseline corresponding to the roller rotation phase point is defined as the unloading span. The ratio of the torque drop amplitude to the unloading span is determined as the torque drop steepness.

[0010] Furthermore, the method for determining the unloaded baseline is as follows: After the unloading phase of this single picking action ends and before the next single picking action begins, a pre-defined unloaded phase interval is extracted. Calculate the arithmetic mean of all real-time drive torque signal values ​​within the no-load phase interval, and define the arithmetic mean as the no-load baseline; The specific process of constructing the feature vector of this single picking action is as follows: The torque rise slope corresponding to the same single picking action is taken as the first dimension component of the feature vector, and the corresponding torque fall steepness is taken as the second dimension component of the feature vector. The two sets of feature components are combined to form a two-dimensional vector, which is defined as the feature vector of this single picking action.

[0011] Furthermore, the pre-calibration method for the harvest quality classifier is as follows: Field harvesting experiments were conducted, and torque maps of the harvesting action of the human hand-like flexible toothed fruit-harvesting roller were collected simultaneously. The skin condition of the pepper fruit corresponding to each harvesting action was recorded frame by frame. Data corresponding to harvesting actions where the pepper skin was not damaged were marked as undamaged harvesting samples, and data corresponding to harvesting actions where the pepper skin was damaged were marked as damaged harvesting samples. Extract the torque ramp slope and torque drop steepness corresponding to all non-destructive harvested samples; On a two-dimensional plane with the torque rise slope as the horizontal axis and the torque fall steepness as the vertical axis, draw a closed polygonal region that encloses all the data corresponding to the non-destructive harvesting samples, and define the closed polygonal region as the non-destructive harvesting feature envelope. The internal region enclosed by the envelope includes the boundary of the envelope itself. The non-destructive picking feature envelope and the corresponding quality judgment rules are stored in the picking quality classifier; The process of classifying the quality of a single harvest based on the non-destructive harvesting characteristic envelope is as follows: If the coordinate point corresponding to the feature vector of this single picking action falls within the inner region of the non-destructive picking feature envelope, then this single picking action is classified as effective non-destructive picking. If the coordinate point corresponding to the feature vector of this single picking action does not fall into the inner region of the non-destructive picking feature envelope, and the shortest distance to the boundary of the non-destructive picking feature envelope is less than the preset distance threshold, then this single picking action is classified as effective minimal-destructive picking. If the coordinate point corresponding to the feature vector of this single picking action does not fall within the inner region of the non-destructive picking feature envelope, and the shortest distance to the boundary of the non-destructive picking feature envelope is greater than or equal to the preset distance threshold, then this single picking action is classified as invalid empty picking.

[0012] Furthermore, the specific method for determining the total amount of pepper input is as follows: The quality type of all single picking actions completed by the human hand-like flexible toothed fruit picking roller within the time window is statistically analyzed, and the effective number of pickings is obtained by adding the number of effective non-damaged pickings to the number of effective slightly damaged pickings. Multiply the effective harvest quantity by the preset average single fruit weight to obtain the quality harvest quantity; Divide the quality harvest quantity by the duration of the time window to obtain the time window harvest efficiency.

[0013] Furthermore, the process of generating continuous time windows along the operation time axis with a fixed sliding step size is as follows: A fixed sliding step size is preset, and the sliding step size is limited to be less than the duration of the time window; The first time window is constructed based on the start time of the harvester operation; Using a fixed sliding step size as the time offset, the machine iteratively shifts backward along the operation time axis until the harvester stops operating, thereby generating several continuous time windows.

[0014] The present invention also provides a device for monitoring the operating efficiency of a self-propelled chili harvester, the device being used to execute a method for monitoring the operating efficiency of a self-propelled chili harvester as described in any of the above claims, comprising: The signal acquisition module is used to acquire the real-time drive torque signal and the roller rotation phase signal of a single complete picking cycle of the human hand-like flexible toothed fruit picking roller, and to establish a motion torque spectrum with a single picking action as the basic unit. The phase division module is used to decompose the action torque spectrum along the phase signal of the drum rotation angle. Based on the zero-crossing characteristics of the torque change rate, a single picking action is decomposed into the loading stage of the progressive insertion of the spring teeth and the unloading stage of the pepper fruit detaching from the stem. The feature fusion module is used to extract the torque ramp-up slope during the loading phase and the torque drop steepness during the unloading phase of this single picking action, and the two are combined to obtain the feature vector of this single picking action. The quality classification module is used to input the feature vector of this single picking action into a pre-calibrated picking quality classifier. Based on the non-destructive picking feature envelope formed by the torque rise slope and torque fall steepness of the non-destructive picking sample, the quality of this single picking action is classified to obtain effective non-destructive picking, effective slightly damaged picking, or invalid empty picking. The efficiency statistics module is used to set a time window, count the number of picking actions of each quality type within the time window, combine the number of effective undamaged picking and the number of effective slightly damaged picking to obtain the effective harvest quantity, calculate the quality harvest quantity by combining the effective harvest quantity with the preset average single fruit quality, and calculate the time window harvest efficiency by combining the quality harvest quantity with the time window duration. The iterative monitoring module is used to generate several continuous time windows along the operation time axis with a fixed sliding step size. It iteratively executes S2 to S5 to complete the calculation of the harvesting efficiency of each time window, arranges the harvesting efficiency of each time window in chronological order, and obtains the continuously changing time series data of harvesting efficiency.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention directly analyzes the torque spectrum of a single picking action and extracts the torque ramp-up slope and torque drop steepness. This method utilizes the zero-crossing characteristic of the torque change rate to decompose each picking action into a loading stage and an unloading stage, allowing the degree of brushing of the plant canopy by the spring teeth and the quality of brittle fracture when the pepper fruit separates to be quantified independently. On this basis, empty swinging actions that fail to produce effective fruit separation are identified and removed from the efficiency statistics, and only actions in which both the torque ramp-up slope and torque drop steepness meet the effective picking conditions are included in the effective fruit picking count. This fundamentally eliminates the interference of invalid actions on the calculated value of operation efficiency, and effectively controls the deviation between the monitoring results and the actual harvest in the field. This invention also solves the problem that existing technologies cannot simultaneously perceive the quality of harvesting actions during efficiency monitoring by synthesizing the torque ramp slope and torque drop steepness into a feature vector for a single harvesting action and comparing it with a feature envelope established based on non-destructive harvesting samples. The quality type of each harvesting action is classified in real time under the constraint of the feature envelope as effective non-destructive harvesting, effective slightly damaged harvesting, or ineffective empty harvesting. This makes the harvesting efficiency obtained within the time window naturally carry analytical information on the integrity of the fruit skin. The resulting operational efficiency monitoring sequence is no longer a single yield rate curve, but a dynamic index sequence that simultaneously reflects the harvesting quantity and harvesting quality, providing a direct basis for real-time adjustment of harvester driving parameters and traceability of operational quality. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall method flow of the present invention; Figure 2 This is a parallel coordinate graph of the torque ramp slope, torque drop steepness, shortest Euclidean distance, and quality classification results of this invention. Figure 3 This is a flowchart of the overall device structure of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0018] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0019] Example: Please see Figures 1-2 The present invention provides a technical solution: A method for monitoring the operating efficiency of a self-propelled chili harvester, comprising the following steps: S1: Obtain the real-time drive torque signal and roller rotation phase signal of a single complete picking cycle of the human hand-like flexible toothed fruit picking roller, and establish a motion torque spectrum with a single picking action as the basic unit. In this embodiment, the core harvesting component of the self-propelled chili harvester is a human-hand-like flexible toothed picking drum. This drum has multiple retractable flexible teeth distributed circumferentially and axially. The harvesting principle involves the flexible teeth gradually extending from the bottom of the drum and inserting into the chili plant canopy. Through a rotating, horizontally lifting motion, the teeth brush the chilies, separating the fruit from the stem. Subsequently, the flexible teeth gradually retract at the top of the drum, completing one harvesting action. The work efficiency monitoring method proposed in this invention is based on a detailed analysis of the mechanical response characteristics exhibited by the human-hand-like flexible toothed picking drum during each harvesting action, thereby achieving online evaluation of harvesting efficiency. The real-time drive torque signal refers to the change signal of the real-time torque value output by the torque sensor installed on the drive transmission chain of the human-hand flexible toothed fruit picking roller, which is used to drive the rotation of the human-hand flexible toothed fruit picking roller. The roller rotation angle phase signal refers to the pulse signal output by the rotation angle encoder installed on the rotation shaft of the human-hand flexible toothed fruit picking roller, which is converted by the vehicle-mounted processing terminal and is a continuous change signal representing the rotation angle position of the human-hand flexible toothed fruit picking roller. The purpose of obtaining the real-time drive torque signal and the roller rotation angle phase signal is to establish a precise correspondence between the mechanical response during the picking process and the rotation position of the human-hand flexible toothed fruit picking roller, so as to provide a data basis for subsequent action disassembly and feature extraction by angle domain. The specific method for obtaining the real-time drive torque signal and the roller rotation angle phase signal is as follows: A torque sensor is installed on the drive transmission chain of the human-hand-like flexible toothed fruit-picking roller, and an angle encoder is installed on the rotating shaft of the human-hand-like flexible toothed fruit-picking roller; the vehicle-mounted processing terminal synchronously collects the analog voltage signal output by the torque sensor at a preset fixed sampling frequency and converts it into the value of the real-time drive torque signal, while simultaneously collecting the rotation angle pulse signal output by the angle encoder; the vehicle-mounted processing terminal accumulates and counts the received rotation angle pulse signals, and multiplies the accumulated count value by the single-pulse angle resolution of the angle encoder to obtain the roller rotation angle phase signal with the sampling time of a preset specific tooth passing through a fixed reference point as the zero point; A complete harvesting cycle is defined as the complete mechanical movement process experienced by a specific pre-set tooth on the human-hand-inspired flexible toothed fruit-picking roller from its initial extension to its complete retraction into the roller. The start and end references of this complete harvesting cycle are determined as follows: a proximity switch sensor is installed on the fixed frame of the human-hand-inspired flexible toothed fruit-picking roller, and the installation position of the proximity switch sensor is defined as a fixed reference point. The sampling moment when the specific tooth rotates with the human-hand-inspired flexible toothed fruit-picking roller and triggers the proximity switch sensor is recorded as the roller rotation start reference. When the specific tooth completes a complete action cycle of its extension mechanism, that is, from its initial extension, reaching its maximum extension, and then completely retracting into the human-hand-inspired flexible toothed fruit-picking roller, and triggers the proximity switch sensor again with the rotation of the roller, the sampling point at the moment of the second trigger is recorded as the roller rotation end reference. The time interval between the roller rotation start reference and the roller rotation end reference is defined as a complete harvesting cycle experienced by the specific tooth. The harvesting torque spectrum refers to a two-dimensional data set indexed by the drum rotation angle phase signal and recorded by the real-time drive torque signal value. This data set completely records the change of drive torque with drum rotation angle phase during a single harvesting action from start to finish. The harvesting torque spectrum is generated as follows: the on-board processing terminal extracts the values ​​of all synchronously collected real-time drive torque signals and drum rotation angle phase signals from the sampling point corresponding to the drum rotation angle start reference to the sampling point corresponding to the drum rotation angle end reference; the value of the drum rotation angle phase signal corresponding to each sampling point is used as the index value, and the value of the real-time drive torque signal corresponding to that sampling point is used as the record value, and the data is stored in a one-to-one correspondence to form a two-dimensional data sequence with drum rotation angle phase as the independent variable and real-time drive torque as the dependent variable. This two-dimensional data sequence is a torque map of the picking action, corresponding one-to-one with the single picking action of the specific tweezers. Thus, the torque map of the picking action, with the single picking action as the basic unit, is established.

[0020] S2: Deconstruct the action torque spectrum along the phase signal of the drum rotation angle, and based on the zero-crossing characteristics of the torque change rate, decompose a single picking action into the loading stage of the progressive insertion of the spring teeth and the unloading stage of the pepper fruit detaching from the stem. In this embodiment, step S2 is to deconstruct the process data of the picking action torque spectrum established in step S1. The purpose is to automatically divide a complete single picking action into two continuous stages with different physical meanings according to the difference in mechanical response characteristics, so as to provide accurate interval boundaries for the stage feature extraction in step S3. Calculate the rate of change of torque of the real-time drive torque signal in the torque spectrum of the picking action relative to the value of the drum rotation angle phase signal. The torque change rate is defined as the derivative of the real-time driving torque signal value in the torque spectrum of the picking action with the value of the drum rotation angle phase signal. It characterizes how fast the driving torque changes with the drum rotation angle. The torque change rate is obtained by the on-board processing terminal calculating the discrete change rate sequence of the two-dimensional data sequence in the torque spectrum of the picking action point by point, using the ratio of the difference between the real-time driving torque signal values ​​between two adjacent sampling points to the difference between the drum rotation angle phase signal values. This discrete change rate sequence is the torque change rate. The zero-crossing characteristic refers to the sign switching phenomenon of the torque change rate changing from a positive value to a negative value or vice versa in its discrete change rate sequence. During a complete single harvesting action, the sign change of the torque change rate reflects the change in the mechanical state of the harvesting action: when the flexible spring teeth gradually insert into the plant canopy and brush it, the driving torque continues to increase, and the torque change rate is positive at this time; when the pepper fruit is detached from the stem, the driving torque drops rapidly from its peak value, and the torque change rate suddenly becomes negative; when the unloading process is completed and the driving torque tends to stabilize, the torque change rate returns to near zero. The vehicle-mounted processing terminal performs zero-crossing detection on the discrete rate of change sequence corresponding to the torque spectrum of the picking action. Based on the zero-crossing characteristics of the torque change rate, it determines the boundary between the loading and unloading stages. The vehicle-mounted processing terminal determines the phase point of the roller rotation angle phase signal corresponding to the torque change rate switching from a positive value to a zero value as the end phase point of the loading stage and the start phase point of the unloading stage. This phase point marks the critical position where the loading process of the flexible spring tooth comb ends and the unloading process of the pepper fruit detaches begins. The on-board processing terminal determines the phase point of the drum rotation angle phase signal when the torque change rate switches from negative to zero as the end phase point of the unloading stage. This phase point marks the critical position where the pepper fruit detachment process ends and the mechanical response of this single picking action basically ends. The loading phase is defined as the phase interval between the starting reference of the roller rotation angle and the ending phase point of the loading phase. Within this interval, the flexible spring teeth are in the process of gradually inserting into the plant canopy and performing combing loading. The driving torque continues to increase as the roller rotation angle phase increases. The unloading phase is defined as the phase interval from the start phase point to the end phase point of the unloading phase. During this interval, the pepper fruit separates from the stem, and the driving torque drops rapidly from its peak position to near the unloaded baseline. The loading and unloading phases together constitute the phase sequence of a single harvesting action, which fully describes the evolution of the mechanical state of a single harvesting action from start to finish.

[0021] S3: Extract the torque ramp-up slope during the loading phase and the torque drop steepness during the unloading phase of this single picking action, and combine the two to obtain the feature vector of this single picking action; In this embodiment, step S3 is based on the interval boundaries of the loading and unloading stages already divided in step S2. The torque spectrum data of the picking action in the two stages are quantized separately, and two independent feature indicators that can characterize the mechanical response characteristics of this single picking action are extracted and combined into a structured feature vector to provide standardized input data for the picking quality classification in step S4. The torque ramp-up slope during the loading phase and the torque drop steepness during the unloading phase of this single harvesting action are extracted separately, with the specific logic as follows: The torque ramp rate is defined as the linear rate of change of the value of the real-time drive torque signal during the loading phase as the value of the roller rotation phase signal increases. Its physical meaning is to characterize the growth intensity of the drive load when the flexible spring teeth are gradually inserted into the plant canopy and brushed. The torque ramp slope is extracted as follows: The on-board processing terminal uses the phase interval corresponding to the loading stage as the data interception range, takes the value of the roller rotation angle phase signal within this interval as the independent variable data column, and takes the value of the real-time drive torque signal as the dependent variable data column. The least squares method is used to perform univariate linear fitting on the corresponding values ​​at the same sampling time in the two sets of data columns to obtain a torque-rotation angle fitting line. The slope of the fitting line is determined as the torque ramp slope of this single picking action. The larger the value of the torque ramp slope, the greater the increase in drive torque within the unit rotation angle interval, reflecting the greater resistance encountered by the flexible spring teeth in the canopy brushing of the plant in this picking action. Specifically, the formula for calculating the torque ramp slope is: in, The torque ramp slope, The first phase interval within the loading phase The value of the drum rotation angle phase signal corresponding to each sampling point. The first phase interval within the loading phase The value of the real-time drive torque signal corresponding to each sampling point. This represents the total number of sampling points within the phase interval during the loading phase. The sampling point number, The value range is from 1 to Integers; Torque ramp slope The physical meaning is the average increase in the value of the real-time drive torque signal within a unit drum rotation angle range. The larger the value, the greater the resistance encountered by the human hand-like flexible spring teeth in the canopy brushing action, and the higher the degree of brushing input. Torque drop steepness is defined as the degree of drastic drop in the real-time drive torque signal from its peak point to the unloaded baseline during the unloading phase. Physically, it characterizes the speed at which the drive load is released when the pepper detaches from the stem. The torque drop steepness is extracted as follows: within the phase interval corresponding to the unloading phase, the on-board processing terminal iterates through the real-time drive torque signal values ​​of all sampling points within that interval. The sampling point with the largest value is identified as the peak point of the real-time drive torque signal, and the peak torque value and the phase point of the roller angle phase signal corresponding to that peak point are recorded. Simultaneously, within the phase interval of the unloading phase, the on-board processing terminal retrieves the sampling point where the real-time drive torque signal value first drops to the unloaded baseline, and records the phase point of the roller angle phase signal corresponding to that sampling point. Specifically, the formula for calculating the torque drop steepness is: in, For the steepness of the torque drop, The peak value of the real-time drive torque signal within the phase interval of the unloading phase is the torque value. For unloaded baseline, This represents the value of the roller rotation angle phase signal corresponding to the peak torque. The value of the drum rotation angle phase signal is the value of the real-time drive torque signal when it first falls back to the no-load baseline. Torque drop steepness The physical meaning is the average rate of decrease of the real-time drive torque signal value within a unit drum rotation angle range. The higher the value, the more decisive the brittle fracture when the pepper fruit detaches from the stem, and the more concentrated the energy release. The on-board processing terminal uses the difference between the peak torque and the no-load baseline as the torque drop amplitude, which reflects the total decrease in drive torque during unloading. The unloading span is used as the angle difference between the phase point of the roller rotation angle phase signal when the torque drops back to the no-load baseline and the phase point of the roller rotation angle phase signal when the torque peak point is reached. The unloading span reflects the range of roller rotation angles experienced during the unloading process. The ratio obtained by dividing the torque drop amplitude by the unloading span is used as the torque drop steepness of this single picking action. The larger the value of the torque drop steepness, the faster the driving torque drops within a unit turning angle range, reflecting that the brittle fracture of the pepper fruit when it detaches from the stem is completed more decisively. The method for determining the unloaded baseline is as follows: The no-load baseline refers to the reference level of the driving torque of the human-hand-like flexible toothed fruit-picking roller in the no-load rotation state. Its function is to serve as a torque reference value to distinguish between the picking load state and the no-load state. The no-load baseline is determined as follows: after the unloading phase of the current single picking action ends and before the start of the next single picking action, the vehicle-mounted processing terminal extracts a pre-set length of no-load phase interval. This no-load phase interval corresponds to the rotation interval on the human-hand-like flexible toothed fruit-picking roller where neither of the two adjacent flexible teeth is in contact with the plant canopy. The vehicle-mounted processing terminal reads the values ​​of the real-time drive torque signals corresponding to all sampling points within the no-load phase interval, calculates the arithmetic mean of these values, and defines the arithmetic mean as the no-load baseline used for this single picking action. The feature vector of this single picking action is defined as a two-dimensional feature data composed of the torque rise slope and torque fall steepness of the same single picking action. Its function is to serve as the input data unit of the picking quality classifier. The feature vector for this single picking action is constructed as follows: the vehicle-mounted processing terminal uses the torque ramp slope obtained from the above extraction process for the same single picking action as the first dimension component of the feature vector, and the torque drop steepness obtained from the above extraction process for the same single picking action as the second dimension component of the feature vector. The first and second dimensions are combined into a two-dimensional vector, and this two-dimensional vector is defined as the feature vector for this single picking action. The feature vector for this single picking action completely characterizes the mechanical response characteristics of this single picking action in the loading and unloading phases.

[0022] S4: Input the feature vector of this single picking action into the pre-calibrated picking quality classifier. Based on the non-destructive picking feature envelope formed by the torque rise slope and torque fall steepness of the non-destructive picking sample, classify the quality of this single picking action to obtain effective non-destructive picking, effective slightly damaged picking, or invalid empty picking. In this embodiment, step S4 is based on the feature vector of the current single picking action obtained in step S3. A pre-established picking quality classifier is used to automatically determine the quality type of the current single picking action. The purpose is to classify each picking action into different quality levels according to its mechanical response characteristics, so as to provide a classification and counting basis for the time window harvesting efficiency statistics in step S5. The pre-calibration method for the harvest quality classifier is as follows: The harvest quality classifier is a classification and judgment model based on pre-calibrated sample data. It internally stores the boundary data of the non-destructive harvest feature envelope and the corresponding quality judgment rules. It is used to output the corresponding single harvest quality type based on the input single harvest action feature vector. The input of the harvest quality classifier is the single harvest action feature vector, and the output is the quality type judgment result of one of the three: effective non-destructive harvest, effective slightly damaged harvest, or invalid empty harvest. The pre-calibration process of the harvest quality classifier is completed before the harvest operation begins, and the specific method is as follows: Under field harvesting test conditions, the self-propelled chili harvester was used to harvest chilies in normal operating condition. At the same time, the on-board processing terminal was used to collect the torque spectrum of the picking action of the human hand-like flexible toothed picking roller during the picking operation. High-speed camera equipment was used to record the chili skin state corresponding to each picking action frame by frame at a frame rate higher than the picking action frequency. The torque spectrum data of the picking action corresponding to the picking action in which the pepper skin was not damaged in the high-speed camera recording was marked as the non-destructive picking sample, and the torque spectrum data of the picking action corresponding to the picking action in which the pepper skin was cracked, squeezed and deformed or damaged in the high-speed camera recording was marked as the destructive picking sample. Extract the torque ramp slope and torque drop steepness corresponding to all non-destructive harvesting samples; each non-destructive harvesting sample corresponds to a set of two-dimensional coordinate data composed of torque ramp slope and torque drop steepness; on a two-dimensional plane with torque ramp slope as the horizontal axis and torque drop steepness as the vertical axis, draw the two-dimensional coordinate data points corresponding to all non-destructive harvesting samples, and use the convex hull algorithm to calculate and draw the smallest closed polygon region that can enclose all the data points corresponding to the non-destructive harvesting samples, and define the closed polygon region as the non-destructive harvesting feature envelope; Specifically, the coordinate point corresponding to the feature vector of this single picking action on a two-dimensional plane with the torque rise slope as the abscissa and the torque fall steepness as the ordinate is denoted as point P, and the coordinates of point P are... Calculate the shortest Euclidean distance from point P to the boundary of the non-destructive harvesting feature envelope. : in, Let P be the shortest Euclidean distance from point P to the boundary of the non-destructive harvesting feature envelope. and These are the first features on the boundary of the non-destructive harvesting feature envelope. The x-coordinate and y-coordinate of the two endpoints of the line segment, where the x-coordinate is the torque ramp-up slope and the y-coordinate is the torque drop steepness; The sequence number of the edge segment of the envelope boundary of the non-destructive harvesting feature. The value range traverses all edge segments of the boundary of the non-destructive harvesting feature envelope; This indicates taking the absolute value. This indicates taking the minimum value among all calculated results; When the foot of the perpendicular from point P to a certain line segment falls outside the endpoint of the line segment, the smaller of the two Euclidean distances from point P to the two endpoints of the line segment is taken as the distance from point P to the line segment. The specific data for the picking action sequence number and the shortest Euclidean distance are shown in Table 1.

[0023] Table 1. Statistical Chart In this data analysis, the torque ramp-up slope, torque drop steepness, and the shortest distance from the combined feature points to the non-destructive harvesting feature envelope of 30 single harvesting actions were systematically evaluated. The data table includes the sequence number of each harvesting action, torque ramp-up slope, torque drop steepness, shortest Euclidean distance, and quality classification results. The preset non-destructive harvesting feature envelope is defined on the two-dimensional feature plane as a rectangular area with a torque ramp-up slope from 10 to 30 and a torque drop steepness from 5 to 15. The preset distance threshold is set to 2.000. Analysis of the data reveals the distribution pattern of the geometric relationship between feature points and the envelope. The shortest Euclidean distances from feature points of picking actions 2, 11, and 26 to the envelope boundary are 1.477, 0.852, and 0.346, respectively, all less than the distance threshold of 2.000, and are therefore classified as effective minimal-loss picking. Among them, the torque ramp slope of action 2 is 8.523, which is 10 below the left boundary of the envelope, and its shortest Euclidean distance is the horizontal distance 1 to the left boundary. 0.477; The torque ramp slope of serial number 11 is 30.852, exceeding the right boundary of the envelope by 30, and the shortest distance is the excess of 0.852; The torque ramp slope of serial number 26 is 9.654, lower than the left boundary, and the torque drop steepness is 12.089, which is inside the envelope. The shortest Euclidean distance is determined by the horizontal difference and is 0.346. Although these points do not fall inside the envelope, the degree of deviation is small, indicating that the mechanical response of the picking action is close to the non-destructive standard. Harvesting actions number 4 and 8 were deemed invalid, with their shortest Euclidean distances reaching 4.070 and 3.110 respectively, far exceeding the distance threshold. The characteristic points of action number 4 were a torque ramp slope of 33.672 and a torque drop steepness of 3.244, exceeding both the right and lower boundaries of the envelope, and its distance was a combination of the excesses in both directions. The characteristic points of action number 8 were a torque ramp slope of 6.890 and a torque drop steepness of 14.337, exceeding only the left boundary, but the horizontal excess reached 3.110, also resulting in a large distance. These two points reflect that the combination of torque ramp slope and torque drop steepness in the harvesting action deviated significantly from the characteristic envelope of non-destructive harvesting, and that effective harvesting may have failed due to factors such as insufficient or excessive resistance, or incomplete detachment of the peppers during the harvesting process. The feature points of the remaining 25 picking actions all fall within the envelope, with the shortest Euclidean distance being 0. They are classified as effective and non-destructive picking. The torque ramp slope of these points is distributed between 11.245 and 29.567, and the torque drop steepness is distributed between 5.789 and 14.756. Overall, they are irregularly scattered within the rectangular area, without obvious clustering trend. For example, the torque ramp slope of number 6 (13.667) and torque drop steepness of number 6 (6.401) are close to the lower left area of ​​the envelope, while the torque ramp slope of number 25 (24.932) and torque drop steepness of number 25 are close to the upper right area of ​​the envelope. There is no simple linear correspondence between the torque ramp slope and the torque drop steepness, indicating that the two features independently reflect the quality of the picking action. Overall, a high torque ramp slope during harvesting often corresponds to significant canopy resistance encountered during the brushing process, similar to a human hand's flexible toothed comb. Conversely, a high torque drop steepness indicates a pronounced brittle fracture when the pepper detaches from the stem. When both characteristics are within the envelope area defined by the undamaged harvesting sample, the harvesting action is considered effective and undamaged. Slight deviation from the envelope boundary is considered effective and slightly damaged. Significant deviation is considered invalid and unharvested. This three-classification method based on envelope distance quantifies the deviation between the mechanical response of the harvesting action and the ideal undamaged state, providing a data foundation for the accurate assessment of harvesting efficiency.

[0024] The inner region of the non-destructive picking feature envelope represents the normal distribution range of the mechanical response characteristics of the picking action when the skin of the pepper fruit is not damaged during the picking process; the vehicle-mounted processing terminal stores the coordinate data of each vertex of the non-destructive picking feature envelope and the corresponding quality judgment rules in the picking quality classifier, and completes the pre-calibration of the picking quality classifier. Effective and non-damaging harvesting is defined as a harvesting action in which the pepper skin is not damaged and the pepper is successfully separated from the stem. Effective and slightly damaged harvesting is defined as a harvesting action in which the pepper skin is slightly damaged but the pepper still has commercial value. Ineffective empty harvesting is defined as a harvesting action in which the flexible spring teeth complete the full telescopic harvesting action but fail to successfully separate the pepper from the stem. The process of classifying the quality of a single harvest based on the non-destructive harvesting characteristic envelope is as follows: The vehicle-mounted processing terminal maps the feature vector of this single picking action onto a two-dimensional plane with the torque climb slope as the horizontal axis and the torque drop steepness as the vertical axis, obtaining a coordinate point. If the coordinate point falls within the inner region of the non-destructive picking feature envelope, it indicates that the mechanical response characteristics of this single picking action are within the normal distribution range of non-destructive picking, and the single picking action is then classified as effective non-destructive picking. If the coordinate point does not fall within the inner region of the non-destructive harvesting feature envelope, the vehicle-mounted processing terminal calculates the shortest Euclidean distance from the coordinate point to all line segments on the boundary of the non-destructive harvesting feature envelope. If the shortest distance is less than a preset distance threshold, it indicates that although the mechanical response characteristics of this single harvesting action deviate from the ideal distribution range of non-destructive harvesting, the degree of deviation is small. In this case, the single harvesting action is classified as effective minimally damaged harvesting. The preset distance threshold is determined based on the statistical distribution of the shortest distance from the coordinate points formed by the torque rise slope and torque fall steepness of all damaged harvesting samples to the boundary of the non-damaged harvesting feature envelope. Specifically, the value is the 20th percentile of the shortest distance corresponding to the damaged harvesting sample. Specifically, the preset distance threshold The formula for determining it is: ,in, The preset distance threshold, This is the set of shortest Euclidean distances from the coordinate points corresponding to all damaged harvesting samples to the boundary of the feature envelope of undamaged harvesting. Each damaged harvesting sample is a data point corresponding to a harvesting action in which the pepper fruit epidermis was damaged, as observed in a field harvesting experiment. This means taking the 20th percentile of the set, i.e., the set After arranging all the values ​​in ascending order, take the value at the 20th percentile position. Preset distance threshold The physical meaning is that in field harvesting experiments, 80% of the damaged harvested samples had a shortest distance greater than the boundary of the non-damaged harvesting characteristic envelope. , As the dividing line between effective minimally invasive harvesting and ineffective empty harvesting; If the coordinate point does not fall within the inner region of the non-destructive harvesting feature envelope and the shortest distance is greater than or equal to the preset distance threshold, it indicates that the mechanical response characteristics of this single harvesting action deviate significantly from the distribution range of non-destructive harvesting, and the single harvesting action is classified as invalid empty harvesting.

[0025] S5: Set a time window, count the number of picking actions of each quality type within the time window, combine the number of effective undamaged picking and the number of effective slightly damaged picking to obtain the effective harvest quantity, calculate the quality harvest quantity with the preset average single fruit quality, and calculate the time window harvest efficiency with the quality harvest quantity and the time window duration. In this embodiment, step S5 is based on the quality classification of each single picking action in step S4. The quality judgment results of all single picking actions that occur within a time period are summarized and statistically analyzed. The discrete quality types of single picking actions are transformed into harvesting efficiency indicators with actual physical meaning, thereby realizing a quantitative evaluation of the harvesting operation effect. The specific method for determining the total amount of pepper input is as follows: The quality type of all single picking actions completed by the human hand-like flexible toothed fruit picking roller within the time window is statistically analyzed, and the effective number of pickings is obtained by adding the number of effective non-damaged pickings to the number of effective slightly damaged pickings. A time window is a statistical interval with a fixed length that is divided along the operation time axis. Its function is to summarize the continuous single harvesting actions in segments according to the time sequence. The duration of the time window is preset according to the operating speed of the harvester and the planting density of the chili plants. The setting range is between 5 seconds and 30 seconds, so as to achieve a balance between statistical stability and response timeliness. After completing the quality classification of each single picking action, the vehicle-mounted processing terminal stores the quality type judgment result of the picking action into the cache in chronological order. When the time window ends, the vehicle-mounted processing terminal traverses all the quality type judgment results stored in the time window and counts the number of effective undamaged picking, the number of effective slightly damaged picking, and the number of invalid empty picking respectively. The effective harvest quantity refers to the total number of harvesting actions within a time window in which the pepper fruit is successfully separated from the stem and the pepper fruit has commercial value. The vehicle-mounted processing terminal adds the number of effective non-damaged harvestings to the number of effective minimally damaged harvestings, and the sum is the effective harvest quantity for that time window. The preset average weight of a single pepper refers to the average weight of a single pepper. It is determined as follows: Before the harvest operation, a preset number of pepper plants are selected in the field to be harvested by random sampling. All mature peppers are picked manually and weighed one by one. The arithmetic mean of all the weights is obtained by adding them together and dividing by the total number of peppers. Quality harvest quantity refers to the total mass of commercially valuable peppers harvested within a time window. The on-board processing terminal multiplies the effective harvest quantity of the time window with the preset average single fruit weight, and the product is the quality harvest quantity of the time window. Time window harvesting efficiency refers to the quality of commercially valuable peppers harvested per unit time. The on-board processing terminal divides the quality harvested quantity of the time window by the duration of the time window, and the result of the division operation is the time window harvesting efficiency. Time window harvesting efficiency comprehensively reflects the output level of the harvester within the time window.

[0026] S6: Generate several continuous time windows along the operation time axis with a fixed sliding step size, and cyclically execute S2 to S5 to complete the window-by-window calculation of the harvesting efficiency. Arrange the harvesting efficiency of each time window in chronological order to obtain continuously changing harvesting efficiency time series data. In this embodiment, step S6 is based on the harvesting efficiency of a single time window obtained in step S5. The statistical method of a single time window is extended to a continuous sliding statistical mechanism along the operation time axis. The purpose is to generate a dynamic change sequence of harvesting efficiency covering the entire harvesting operation process, so as to provide real-time and continuous harvesting efficiency feedback information for the harvester driver or the onboard control system. The process of generating continuous time windows along the operation time axis with a fixed sliding step size is as follows: The operation time axis refers to a one-dimensional time coordinate axis with the moment when the harvester starts the field harvesting operation as the origin and extends along the actual operation time. It is used to mark the time position of each time window in the entire harvesting operation process. The harvester operation start time is defined as the moment when the self-propelled chili harvester starts the chili harvesting operation in the field. This moment is automatically determined by the on-board processing terminal by detecting the first time that the drive torque signal of the human hand-like flexible tooth picking roller continuously exceeds the preset time of the no-load baseline. A fixed sliding step size is preset, and the sliding step size is limited to be less than the duration of the time window; A fixed sliding step size refers to a fixed time interval between the start times of two adjacent time windows. The value of the fixed sliding step size is less than the duration of a single time window, and the ratio of the fixed sliding step size to the duration of the time window determines the degree of data overlap between adjacent time windows. The fixed sliding step size is preset before the harvesting operation begins. A typical fixed sliding step size is between one-quarter and one-half of the duration of the time window, which allows for some data overlap between adjacent time windows, thereby ensuring the continuity and smoothness of the harvesting efficiency time series data on the time axis. A continuous time window refers to a series of time windows generated sequentially along the operation time axis at fixed sliding step intervals. The start times of adjacent time windows differ by a fixed sliding step, and there is a partial time overlap between adjacent time windows. The specific process of generating a continuous time window is as follows: the on-board processing terminal takes the start time of the harvester operation as the start time of the first time window, and the end time of the first time window is its start time plus the duration of the time window. The vehicle-mounted processing terminal adds a fixed sliding step to the start time of the first time window to obtain the start time of the second time window, adds the duration of the second time window to the start time of the second time window to obtain the end time of the second time window, and so on. Using a fixed sliding step as the time offset, the time windows are generated iteratively along the operation time axis until the harvest operation stops, thereby generating several consecutive time windows. For each generated time window, the vehicle-mounted processing terminal sequentially executes steps S2 to S5 to complete the stage division, feature extraction, quality classification and efficiency statistics of all single picking actions within the time window, and obtains the time window harvesting efficiency corresponding to the time window. Harvesting efficiency time series data refers to a time-varying data sequence formed by arranging the harvesting efficiency calculated for each time window in chronological order of the start time of the corresponding time window on the operation time axis; after the vehicle-mounted processing terminal completes the harvesting efficiency calculation for each time window, it stores the harvesting efficiency value of the time window and the start time of the time window as a set of time series data points into the harvesting efficiency time series data sequence. As harvesting operations continue and continuous time windows are generated, the harvesting efficiency time-series data sequence is continuously updated and extended along the operation time axis, forming a continuous curve that reflects the dynamic changes in the harvesting efficiency of the self-propelled chili harvester over time, providing data support for real-time monitoring of harvesting operation results and online adjustment of operation parameters.

[0027] Please see Figure 3 The present invention also provides a self-propelled chili harvester operation efficiency monitoring device, the device being used to execute the self-propelled chili harvester operation efficiency monitoring method described in any of the above claims, comprising: The signal acquisition module is used to acquire the real-time drive torque signal and the roller rotation phase signal of a single complete picking cycle of the human hand-like flexible toothed fruit picking roller, and to establish a motion torque spectrum with a single picking action as the basic unit. The phase division module is used to decompose the action torque spectrum along the phase signal of the drum rotation angle. Based on the zero-crossing characteristics of the torque change rate, a single picking action is decomposed into the loading stage of the progressive insertion of the spring teeth and the unloading stage of the pepper fruit detaching from the stem. The feature fusion module is used to extract the torque ramp-up slope during the loading phase and the torque drop steepness during the unloading phase of this single picking action, and the two are combined to obtain the feature vector of this single picking action. The quality classification module is used to input the feature vector of this single picking action into a pre-calibrated picking quality classifier. Based on the non-destructive picking feature envelope formed by the torque rise slope and torque fall steepness of the non-destructive picking sample, the quality of this single picking action is classified to obtain effective non-destructive picking, effective slightly damaged picking, or invalid empty picking. The efficiency statistics module is used to set a time window, count the number of picking actions of each quality type within the time window, combine the number of effective undamaged picking and the number of effective slightly damaged picking to obtain the effective harvest quantity, calculate the quality harvest quantity by combining the effective harvest quantity with the preset average single fruit quality, and calculate the time window harvest efficiency by combining the quality harvest quantity with the time window duration. The iterative monitoring module is used to generate several continuous time windows along the operation time axis with a fixed sliding step size. It iteratively executes S2 to S5 to complete the calculation of the harvesting efficiency of each time window, arranges the harvesting efficiency of each time window in chronological order, and obtains the continuously changing time series data of harvesting efficiency.

[0028] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0029] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0030] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0031] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for monitoring the operating efficiency of a self-propelled chili harvester, characterized in that, The specific steps include: S1: Obtain the real-time drive torque signal and roller rotation phase signal of a single complete picking cycle of the human hand-like flexible toothed fruit picking roller, and establish a motion torque spectrum with a single picking action as the basic unit. S2: Deconstruct the action torque spectrum along the phase signal of the drum rotation angle, and based on the zero-crossing characteristics of the torque change rate, decompose a single picking action into the loading stage of the progressive insertion of the spring teeth and the unloading stage of the pepper fruit detaching from the stem. S3: Extract the torque ramp-up slope during the loading phase and the torque drop steepness during the unloading phase of this single picking action, and combine the two to obtain the feature vector of this single picking action; S4: Input the feature vector of this single picking action into the pre-calibrated picking quality classifier. Based on the non-destructive picking feature envelope formed by the torque rise slope and torque fall steepness of the non-destructive picking sample, classify the quality of this single picking action to obtain effective non-destructive picking, effective slightly damaged picking, or invalid empty picking. S5: Set a time window, count the number of picking actions of each quality type within the time window, combine the number of effective undamaged picking and the number of effective slightly damaged picking to obtain the effective harvest quantity, calculate the quality harvest quantity with the preset average single fruit quality, and calculate the time window harvest efficiency with the quality harvest quantity and the time window duration. S6: Generate several continuous time windows along the operation time axis with a fixed sliding step size. Repeatedly execute S2 to S5 to complete the window-by-window calculation of the harvesting efficiency. Arrange the harvesting efficiency of each time window in chronological order to obtain continuously changing harvesting efficiency time series data.

2. The method for monitoring the operating efficiency of a self-propelled chili harvester according to claim 1, characterized in that: The real-time drive torque signal output by the torque sensor and the rotation pulse signal output by the rotation encoder are synchronously acquired at a fixed sampling frequency, and the rotation pulse signal is converted into the drum rotation phase signal. The sampling time when a specific spring tooth on the human hand-like flexible spring tooth fruit picking roller passes through a fixed reference point is used as the starting reference for the roller's rotation angle, and the sampling time when the specific spring tooth completes one complete telescopic picking cycle and passes through the fixed reference point again is used as the ending reference for the roller's rotation angle. The real-time drive torque signal and the drum rotation angle phase signal, which are synchronously collected within the phase interval of the drum rotation angle start reference and the drum rotation angle end reference, are associated and stored point by point to obtain the picking action torque map corresponding to a single picking action.

3. The method for monitoring the operating efficiency of a self-propelled chili harvester according to claim 2, characterized in that: Calculate the rate of change of torque of the real-time drive torque signal in the torque spectrum of the picking action relative to the value of the drum rotation angle phase signal. The phase point of the drum rotation angle phase signal corresponding to the change rate of torque from a positive value to a zero value is defined as the end phase point of the loading phase and the start phase point of the unloading phase. The phase point of the drum rotation angle phase signal corresponding to the change rate of torque from negative to zero is defined as the end phase point of the unloading phase. The interval between the starting reference of the drum rotation angle and the ending phase point of the loading stage is divided into the loading stage; The interval between the start phase point and the end phase point of the unloading phase is defined as the unloading phase.

4. The method for monitoring the operating efficiency of a self-propelled chili harvester according to claim 3, characterized in that: The torque ramp-up slope during the loading phase and the torque drop steepness during the unloading phase of this single harvesting action are extracted separately, with the specific logic as follows: Using the value of the drum rotation phase signal during the loading stage as the independent variable and the value of the real-time drive torque signal as the dependent variable, the corresponding data of the two are fitted with least squares, and the slope of the fitted line is determined as the torque climb slope of the picking action. Extract the peak torque value corresponding to the peak point of the real-time drive torque signal during the unloading phase, the roller rotation phase point corresponding to the peak torque value, and the roller rotation phase point corresponding to the torque value falling back to the no-load baseline. The difference between the peak torque and the unloaded baseline is used as the torque drop amplitude. The angle difference between the peak torque corresponding to the roller rotation phase point and the torque drop to the unloaded baseline corresponding to the roller rotation phase point is defined as the unloading span. The ratio of the torque drop amplitude to the unloading span is determined as the torque drop steepness.

5. The method for monitoring the operating efficiency of a self-propelled chili harvester according to claim 4, characterized in that: The method for determining the unloaded baseline is as follows: After the unloading phase of this single picking action ends and before the next single picking action begins, a pre-defined unloaded phase interval is extracted. Calculate the arithmetic mean of all real-time drive torque signal values ​​within the no-load phase interval, and define the arithmetic mean as the no-load baseline; The specific process of constructing the feature vector of this single picking action is as follows: The torque rise slope corresponding to the same single picking action is taken as the first dimension component of the feature vector, and the corresponding torque fall steepness is taken as the second dimension component of the feature vector. The two sets of feature components are combined to form a two-dimensional vector, which is defined as the feature vector of this single picking action.

6. The method for monitoring the operating efficiency of a self-propelled chili harvester according to claim 1, characterized in that: The pre-calibration method for the harvest quality classifier is as follows: Field harvesting experiments were conducted, and torque maps of the harvesting action of the human hand-like flexible toothed fruit-harvesting roller were collected simultaneously. The skin condition of the pepper fruit corresponding to each harvesting action was recorded frame by frame. Data corresponding to harvesting actions where the pepper skin was not damaged were marked as undamaged harvesting samples, and data corresponding to harvesting actions where the pepper skin was damaged were marked as damaged harvesting samples. Extract the torque ramp slope and torque drop steepness corresponding to all non-destructive harvested samples; On a two-dimensional plane with the torque rise slope as the horizontal axis and the torque fall steepness as the vertical axis, draw a closed polygonal region that encloses all the data corresponding to the non-destructive harvesting samples, and define the closed polygonal region as the non-destructive harvesting feature envelope. The internal region enclosed by the envelope includes the boundary of the envelope itself. The non-destructive picking feature envelope and the corresponding quality judgment rules are stored in the picking quality classifier; The process of classifying the quality of a single harvest based on the non-destructive harvesting characteristic envelope is as follows: If the coordinate point corresponding to the feature vector of this single picking action falls within the inner region of the non-destructive picking feature envelope, then this single picking action is classified as effective non-destructive picking. If the coordinate point corresponding to the feature vector of this single picking action does not fall into the inner region of the non-destructive picking feature envelope, and the shortest distance to the boundary of the non-destructive picking feature envelope is less than the preset distance threshold, then this single picking action is classified as effective minimal-destructive picking. If the coordinate point corresponding to the feature vector of this single picking action does not fall within the inner region of the non-destructive picking feature envelope, and the shortest distance to the boundary of the non-destructive picking feature envelope is greater than or equal to the preset distance threshold, then this single picking action is classified as invalid empty picking.

7. The method for monitoring the operating efficiency of a self-propelled chili harvester according to claim 1, characterized in that: The specific method for determining the total amount of pepper input is as follows: The quality type of all single picking actions completed by the human hand-like flexible toothed fruit picking roller within the time window is statistically analyzed, and the effective number of pickings is obtained by adding the number of effective non-damaged pickings to the number of effective slightly damaged pickings. Multiply the effective harvest quantity by the preset average single fruit weight to obtain the quality harvest quantity; Divide the quality harvest quantity by the duration of the time window to obtain the time window harvest efficiency.

8. The method for monitoring the operating efficiency of a self-propelled chili harvester according to claim 1, characterized in that: The process of generating continuous time windows along the operation time axis with a fixed sliding step size is as follows: A fixed sliding step size is preset, and the sliding step size is limited to be less than the duration of the time window; The first time window is constructed based on the start time of the harvester operation; Using a fixed sliding step size as the time offset, the machine iteratively shifts backward along the operation time axis until the harvester stops operating, thereby generating several continuous time windows.

9. A device for monitoring the operating efficiency of a self-propelled chili harvester, characterized in that: The device is used to perform a method for monitoring the operating efficiency of a self-propelled chili harvester as described in any one of claims 1-8, comprising: The signal acquisition module is used to acquire the real-time drive torque signal and the roller rotation phase signal of a single complete picking cycle of the human hand-like flexible toothed fruit picking roller, and to establish a motion torque spectrum with a single picking action as the basic unit. The phase division module is used to decompose the action torque spectrum along the phase signal of the drum rotation angle. Based on the zero-crossing characteristics of the torque change rate, a single picking action is decomposed into the loading stage of the progressive insertion of the spring teeth and the unloading stage of the pepper fruit detaching from the stem. The feature fusion module is used to extract the torque ramp-up slope during the loading phase and the torque drop steepness during the unloading phase of this single picking action, and the two are combined to obtain the feature vector of this single picking action. The quality classification module is used to input the feature vector of this single picking action into a pre-calibrated picking quality classifier. Based on the non-destructive picking feature envelope formed by the torque rise slope and torque fall steepness of the non-destructive picking sample, the quality of this single picking action is classified to obtain effective non-destructive picking, effective slightly damaged picking, or invalid empty picking. The efficiency statistics module is used to set a time window, count the number of picking actions of each quality type within the time window, combine the number of effective undamaged picking and the number of effective slightly damaged picking to obtain the effective harvest quantity, calculate the quality harvest quantity by combining the effective harvest quantity with the preset average single fruit quality, and calculate the time window harvest efficiency by combining the quality harvest quantity with the time window duration. The iterative monitoring module is used to generate several continuous time windows along the operation time axis with a fixed sliding step size. It iteratively executes S2 to S5 to complete the calculation of the harvesting efficiency of each time window, arranges the harvesting efficiency of each time window in chronological order, and obtains the continuously changing time series data of harvesting efficiency.

Images