A grain-treading inspection robot integrating machine vision and edge computing and an inspection method

Abstract

The application provides a grain-treading inspection robot integrating machine vision and edge computing and an inspection method, comprising a main control box, a support structure, a driving system, a plug-in storage detector, an image recognizer, an edge computing board, an indoor positioning module and an environmental gas detector; the plug-in storage detector is used for inserting into a warehouse stack to collect temperature and humidity data; the image recognizer is used for collecting surface images of the warehouse stack; the indoor positioning module is used for obtaining the position and orientation of the robot; the environmental gas detector is used for detecting gas components in the warehouse environment; the edge computing board receives remote instructions, processes data of the plug-in storage detector, identifies images of the image recognizer, plans an inspection path and sends driving instructions to the driving system; the application realizes multi-point and distributed sensing of the warehouse environment, breaks through the limitation of limited coverage of traditional fixed-point monitoring, thereby improving environmental information monitoring and pest monitoring effect and realizing more efficient warehouse prevention and control.

Description

Technical Field

[0001] This invention belongs to the field of storage technology, and in particular relates to a grain inspection robot and inspection method that integrates machine vision and edge computing. Background Technology

[0002] In the management of materials vital to the national economy and people's livelihoods, such as key categories like grains, medicinal herbs, and tobacco, ensuring that the storage environment remains stable and suitable over the long term is of decisive significance. Precise regulation of temperature and humidity in storage spaces, and monitoring and dynamic control of key gas components such as oxygen and carbon dioxide, can effectively inhibit biological attacks such as pests and molds, maximizing the preservation of the use and economic value of the stored goods. Loss of control over any of these core environmental parameters can disrupt the inherent balance of the storage system, triggering a chain reaction from localized deterioration to large-scale damage. Therefore, establishing and operating a precise, reliable, and traceable refined environmental management system has become the cornerstone and fundamental work for ensuring the safety of reserves.

[0003] Current methods for monitoring the storage environment face a dual dilemma in practical applications: limited coverage and insufficient sensing capabilities. This leads to two extremely urgent practical needs. First, there is a need for monitoring methods that can move freely across the entire grain pile surface, rather than being confined to pre-set walkways. Large grain warehouses have a large spatial scale and diverse regional distribution, with a vast and continuous grain pile surface. However, existing monitoring methods often rely on fixed walkways for point-to-point or line-by-line inspections, leaving large areas of the grain surface as unreachable blind spots. To truly capture "hot spots" or anomalies caused by uneven airflow distribution, localized heat sources, and differences in sealing, it is necessary to overcome the physical limitations of the walkways and achieve free-moving monitoring of the entire grain surface without blind spots or interference, without disrupting the flatness of the grain pile surface. Second, while manually treading the grain surface can effectively identify potential problems, it relies on experience and is time-consuming and labor-intensive. By treading the grain surface, storage personnel can judge the density of the grain layer by feel, and promptly detect early signs of spoilage such as clumping, overheating, mold, condensation, and even voids. However, the inspection of grain surfaces relies entirely on manpower, which is extremely time-consuming and labor-intensive. Furthermore, the trampling itself can disrupt the flatness of the grain surface, making it difficult to implement frequently and comprehensively. When dealing with large-scale grain warehouses, the timeliness and comprehensiveness of the hazard inspection cannot be guaranteed.

[0004] The two realities mentioned above highlight the fundamental deficiencies of traditional monitoring models in terms of completeness, timeliness, and cost-effectiveness. This makes it imperative to develop a new, refined management method that can move freely along the entire grain surface without damaging its flatness, and can replace manual methods to deeply perceive potential grain conditions and risks. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a grain inspection robot and inspection method that integrates machine vision and edge computing, thereby improving the effectiveness of environmental information monitoring and pest monitoring, and achieving more efficient warehouse pest control.

[0006] Note that the description of these objectives does not preclude the existence of other objectives. One aspect of the invention does not require achieving all of the above objectives. Objectives other than those described above can be extracted from the description, drawings, and claims.

[0007] The present invention achieves the above-mentioned technical objectives through the following technical means.

[0008] A grain inspection robot integrating machine vision and edge computing includes a main control box, a support structure, a drive system, an insertable storage detector, an image recognizer, an edge computing board, an indoor positioning module, and an environmental gas detector.

[0009] The main control box is fixed on the support structure, the drive system is connected to the support structure and supports the support structure, and the drive system is used to drive the robot to move and turn.

[0010] The insertable storage detector is installed on the main control box and is used to insert into the storage stack to collect temperature and humidity data.

[0011] The image recognition device is slidably connected to the bottom of the main control box and is used to collect images of the surface of the warehouse stacks;

[0012] The indoor positioning module is installed inside the main control box and is electrically connected to the edge computing board to obtain the robot's position and orientation.

[0013] The ambient gas detector is installed inside the main control box and electrically connected to the edge computing board to detect the gas composition in the storage environment.

[0014] The edge computing board is located inside the main control box and serves as the main control unit. It is electrically connected to the indoor positioning module, the ambient gas detector, the drive system, the insertable storage detector, and the image recognition device. It is used to receive remote commands, process data from the insertable storage detector, recognize images from the image recognition device, plan inspection paths, and send drive commands to the drive system.

[0015] The above solution also includes a remote monitoring system;

[0016] The remote monitoring system is connected to the main control box and is used to acquire real-time images in manual remote control mode and send them directly to the remote control center.

[0017] In the above scheme, the support structure includes a driven wheel bracket and a drive wheel bracket, which are located on the left and right sides of the robot and are arranged parallel to each other.

[0018] The driven wheel bracket includes: a cylinder wheel connecting rod, two connecting rods, two right-angle connectors, and two push rod rotating connectors; wherein the two connecting rods are parallel and spaced apart, one end of each connecting rod is connected to one end of the cylinder wheel connecting rod through a right-angle connector, and the other end of each connecting rod is connected to a push rod rotating connector;

[0019] The drive wheel bracket includes: another cylindrical wheel connecting rod, two push rod motors, a support rod, two push rod right-angle connectors, and two support rod connectors; wherein the two push rod motors are arranged in parallel, and the housing of each push rod motor is connected to one end of the other cylindrical wheel connecting rod through a push rod right-angle connector; the two ends of the support rod are fixedly connected to the housing of the push rod motor on the same side through support rod connectors; the output shaft of the push rod motor is hinged to the push rod rotating connector, and the hinge holes of the two push rod rotating connectors are parallel to each other and located in the same plane.

[0020] Furthermore, one end of the push rod rotating connector is hinged to the output shaft of the push rod motor, and the other end is fixedly connected to the connecting rod; the housing of the push rod motor is fixed to the drive wheel bracket through the push rod right-angle connector;

[0021] When turning, only one of the push rod motors operates: when turning left, the push rod of the push rod motor on the right extends outward, while the push rod motor on the left remains stationary. The driven wheel rotates around the hinge point on the left, i.e., the connection point between the left push rod rotating connector and the output shaft of the push rod motor, to achieve overall deflection of the driven wheel; the opposite is true when turning right.

[0022] In the above scheme, the drive system includes two roller assemblies, namely a drive wheel and a driven wheel;

[0023] Each of the driving wheel and the driven wheel includes: a lightweight roller, an anti-slip layer, a roller support, a bearing, and a bearing fixing member; wherein, the anti-slip layer is provided on the outer circumference of the lightweight roller, and the interior of the lightweight roller is fixedly connected to the roller support; the roller support has a shaft hole at its center, the outer ring of the bearing is fixedly connected to the shaft hole of the roller support, and the inner ring of the bearing is fixedly connected to the bearing fixing member; the bearing fixing member is fixedly connected to the roller connecting rod; through the bearing, the lightweight roller can rotate freely relative to the roller connecting rod;

[0024] The drive wheel further includes: a motor support and a rolling drive motor; wherein, the motor support is sleeved on the roller connecting rod in the drive wheel bracket; one end of the motor support is fixedly connected to the housing of the rolling drive motor, a drive gear is provided on the output shaft of the rolling drive motor, and a driven gear is provided on the roller support of the drive wheel, the drive gear and the driven gear mesh with each other, and when the rolling drive motor rotates, the lightweight roller of the drive wheel is driven to rotate through the gear meshing.

[0025] In the above scheme, the insertable storage detector includes a rack and pinion stop, a rack, a descent drive motor, and a temperature and humidity detection probe;

[0026] The rack limiting component is a housing structure with a longitudinal sliding groove, which is fixedly installed on the main control box; the rack is embedded in the sliding groove of the rack limiting component and can slide up and down along the sliding groove; the housing of the descent drive motor is fixed to the upper end of the rack limiting component, and a descent drive gear is installed on the output shaft of the descent drive motor; a rack gear structure is provided on one side of the rack, which meshes with the descent drive gear.

[0027] The temperature and humidity detection probe is fixedly connected to the lower end of the rack;

[0028] When the descent drive motor rotates forward, it drives the rack to slide downward through gear engagement, causing the temperature and humidity detection probe to insert into the surface of the storage stack; when the descent drive motor rotates in reverse, it drives the rack to slide upward, causing the temperature and humidity detection probe to leave the surface of the storage stack.

[0029] Furthermore, the main control box is also equipped with a motor drive board, which is electrically connected to the descent drive motor and is used to monitor the power of the descent drive motor and send the power value to the edge computing board in real time. The edge computing board is used to determine whether the power value exceeds a preset threshold, identify the clumping or caking of the surface of the storage stack based on the determination result, and trigger an abnormal warning when the threshold is exceeded.

[0030] Furthermore, the main control box is divided into a lower main control box layer, a middle main control box layer, and an upper main control box layer arranged from top to bottom;

[0031] The lower layer of the main control box is fixedly equipped with a battery box, a motor drive board and a battery box baffle. The battery box is embedded in the bottom of the lower layer of the main control box. The battery box baffle is connected to the lower layer of the main control box by screws and stops the battery box.

[0032] The edge computing board, the ambient gas detector, and the indoor positioning module are fixedly installed in the middle layer of the main control box; the ambient gas detector is electrically connected to the edge computing board and is used to detect the gas composition in the storage environment and send the detection data to the edge computing board.

[0033] The top of the main control box is fixedly connected to the bottom of the remote monitoring system.

[0034] Furthermore, the bottom of the main control box is provided with a symmetrical longitudinal rod structure, and the rod structure has a long slot along its length; the image recognizer is provided with plug-in parts on both sides, the plug-in parts are nested outside the rod structure and can slide up and down along the rod structure, and the height is fixed by tightening the locking screw through the plug-in parts and the long slot to change the field of view height of the camera.

[0035] An inspection method for a grain inspection robot based on integrated machine vision and edge computing includes the following steps:

[0036] Step S1: The remote operation center sends an automatic inspection command, and the edge computing board initializes and calibrates the robot after receiving the command;

[0037] Step S2: The edge computing board obtains the robot's position and orientation in real time through the indoor positioning module, and controls the drive system to make the robot move autonomously to the preset task point;

[0038] Step S3: After reaching the task point, the edge computing board controls the ambient gas detector to collect ambient gas data, controls the insertable storage detector to insert into the storage stack to collect temperature and humidity data and monitors the power of the descending drive motor to determine the clumping situation, and controls the image recognizer to collect images of the storage stack surface.

[0039] Step S4: The edge computing board runs the pre-deployed YOLO pest identification model to identify the collected images and record the types and quantities of pests;

[0040] Step S5: When any collected data exceeds the set threshold, the edge computing board sends an early warning message and its current location to the remote monitoring center;

[0041] Step S6: After completing the current task point, the robot moves to the next task point and repeats steps S3 to S5 until all task points have been inspected.

[0042] Step S7: The robot returns to the preset standby point, stores all detection data locally, and uploads it to the cloud.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] 1. This invention integrates a main control box, support architecture, drive system, plug-in storage detector, image recognition device, and edge computing.

[0045] The robot integrates the control panel, indoor positioning module, and environmental gas detector into one unit. The drive system is connected to the supporting structure and lifts the robot as a whole. Combined with the lightweight wheel structure, this allows the robot to move freely on the surface of the warehouse stacks without disrupting the flatness of the stacks. The indoor positioning module acquires the robot's position and orientation in real time. The edge computing board, as the main control unit, processes the positioning data, gas composition, temperature and humidity, and image information, and plans the inspection path. This enables multi-point, distributed perception of the warehouse environment, breaking through the limitations of traditional fixed-point monitoring with its limited coverage.

[0046] 2. The addition of the remote monitoring system in this invention enables the robot to have a manual remote control mode in addition to automatic inspection.

[0047] Operators can view real-time warehouse footage through a remote control center, allowing them to understand the on-site situation without entering the warehouse area. This provides an auxiliary means for emergency response and does not interfere with the automatic inspection mode.

[0048] 3. The support structure of this invention adopts a symmetrical arrangement of the driven wheel bracket and the drive wheel bracket, and the output shaft of the push rod motor and the push rod...

[0049] The rotating connector is hinged, requiring only a single push rod action to achieve overall deflection of the driven wheel, resulting in a simple steering structure. Both the drive and driven wheels utilize lightweight cylindrical wheels with anti-slip layers, and the cooperation between the bearings and the cylindrical wheel connecting rod ensures flexible wheel rotation. The drive wheel's rolling drive motor is fixed to the support rod via a motor support component, and power is transmitted through gear meshing, resulting in a compact structure. This design facilitates the robot's movement and steering on soft or uneven grain surfaces.

[0050] 4. The insertion-type storage detector of this invention adopts a rack and pinion lifting mechanism, and the descent drive motor drives the temperature and humidity sensor through a gear and rack.

[0051] The detection probe is inserted vertically into the surface of the stack, directly acquiring temperature and humidity data from inside the stack, supplementing the limitations of surface monitoring alone. The motor drive board monitors the power of the drop drive motor in real time and sends it to the edge computing board. The edge computing board automatically determines whether the stack has abnormalities such as clumping or compaction based on whether the power exceeds a preset threshold. This method provides an objective basis for judgment, replacing manual experience-based grain treading, and can trigger an alert when an anomaly is detected.

[0052] 5. The main control box of this invention is divided into three layers: upper, middle, and lower, which respectively house the battery box, motor drive board, edge computing board, and ring.

[0053] The system includes an ambient gas detector, an indoor positioning module, and a remote monitoring system, facilitating maintenance and quick battery replacement (a battery compartment cover prevents it from falling off). The image recognition device slides through a connector onto a long slot structure, allowing manual adjustment and locking of the camera height to accommodate different stacking heights or field-of-view requirements, thus helping to capture clear images of pests.

[0054] 6. The automatic inspection method provided by this invention includes: automatic positioning and navigation, autonomous movement to the task point, and collection of ambient gas.

[0055] The method involves inserting data into a stack to obtain temperature, humidity, and clumping information; capturing images of the stack surface; running a pre-trained YOLO pest identification model on an edge computing board; recording pest types and quantities; sending warnings and location data to a remote location when thresholds are exceeded; and storing all data locally before uploading it to the cloud. This method performs image recognition and anomaly detection locally on the robot, eliminating the need to upload large amounts of raw images to the cloud. This reduces network bandwidth usage, shortens data transmission links, automates the inspection process, and ensures data traceability, thus reducing the workload of manual inspections.

[0056] Note that the description of these effects does not preclude the existence of other effects. One aspect of the invention does not necessarily have to have all of the above.

[0057] The effects described above are obvious from the description, drawings, claims, etc. Attached Figure Description

[0058] Figure 1 This is a three-dimensional schematic diagram of a grain inspection robot integrating machine vision and edge computing according to an embodiment of the present invention.

[0059] Figure 2 This is a three-dimensional schematic diagram of the lower layer of the main control box according to an embodiment of the present invention.

[0060] Figure 3 This is a three-dimensional schematic diagram of the middle layer of the main control box according to an embodiment of the present invention.

[0061] Figure 4 This is a three-dimensional schematic diagram of the upper layer of the main control box according to an embodiment of the present invention.

[0062] Figure 5 This is a three-dimensional schematic diagram of the support structure according to one embodiment of the present invention.

[0063] Figure 6 This is a three-dimensional schematic diagram of a drive system according to an embodiment of the present invention.

[0064] Figure 7 This is a schematic diagram of the cooperation between the support structure and the drive system according to one embodiment of the present invention.

[0065] Figure 8This is a three-dimensional schematic diagram of an insertable storage detector according to an embodiment of the present invention.

[0066] Figure 9 This is a three-dimensional schematic diagram of a remote monitoring system according to an embodiment of the present invention.

[0067] Figure 10 This is a three-dimensional schematic diagram of the rotational engagement of a remote monitoring system according to an embodiment of the present invention.

[0068] Figure 11 This is a diagram showing the single-target detection effect of the YOLO26n model for warehouse pests according to one embodiment of the present invention.

[0069] Figure 12 This is a diagram showing the multi-target detection effect of the YOLO26n model for warehouse pests according to one embodiment of the present invention.

[0070] Figure 13 This is a diagram illustrating the pest detection effect of the YOLO26n model in a complex storage scenario according to an embodiment of the present invention.

[0071] Figure 14 This is a diagram showing the pest detection effect of the YOLO26n model in a real-world warehouse environment according to one embodiment of the present invention.

[0072] In the diagram, 1. Main control box, 2. Support structure, 3. Drive system, 4. Insert-type storage detector, 5. Remote monitoring system, 6. Image recognition device, 111. Lower layer of main control box, 112. Battery box, 113. Motor drive board, 114. Battery box baffle, 121. Middle layer of main control box, 122. Ambient gas detector, 123. Indoor positioning module, 124. Edge computing board, 131. Upper layer of main control box, 201. Piston wheel connecting rod, 202. Connecting rod, 203. Push rod motor, 204. Support rod, 205. Right angle connector, 206. Push rod straight 207. Angle connector, 208. Push rod rotating connector, 301. Support rod connector, 302. Lightweight roller, 303. Anti-slip layer, 304. Roller support, 305. Bearing, 306. Bearing fixing component, 307. Rolling drive motor, 401. Rack limit component, 402. Rack, 403. Descent drive motor, 404. Temperature and humidity detection probe, 501. Camera fixing component, 502. Camera upper cover, 503. Camera lower cover, 504. Rotating servo motor one, 505. Camera rotating component, 506. Rotating servo motor two. Detailed Implementation

[0073] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0074] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "front," "rear," "left," "right," "upper," "lower," "axial," "radial," "vertical," "horizontal," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0075] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0076] like Figure 1 As shown, this is a preferred embodiment of the grain inspection robot integrating machine vision and edge computing according to the present invention. The grain inspection robot integrating machine vision and edge computing includes a main control box 1, a support structure 2, a drive system 3, an insertable storage detector 4, an image recognizer 6, an edge computing board 124, an indoor positioning module 123, and an environmental gas detector 122.

[0077] The main control box 1 is fixed on the support structure 2. The drive system 3 is connected to the support structure 2 and supports the support structure 2. The drive system 3 is used to drive the robot to move and turn. The insertable storage detector 4 is installed on the main control box 1 and is used to insert into the storage stack to collect temperature and humidity data. The image recognizer 6 is slidably connected to the bottom of the main control box 1 and is used to collect images of the storage stack surface. The indoor positioning module 123 is installed inside the main control box 1 and is electrically connected to the edge computing board 124 to obtain the robot's position and orientation. The ambient gas detector 122 is installed inside the main control box 1 and is electrically connected to the edge computing board 124 to detect the gas composition in the storage environment. The edge computing board 124 is set inside the main control box 1 as the main control unit. It is electrically connected to the indoor positioning module 123, the ambient gas detector 122, the drive system 3, the insertable storage detector 4, and the image recognizer 6. It is used to receive remote commands, process the data of the insertable storage detector 4, recognize the images of the image recognizer 6, plan the inspection path, and send drive commands to the drive system 3. Specifically, the main control box 1 is fixed above the support structure 2, and the drive system 3 is connected to the support structure 2 via a pin to achieve a rotation effect. The insertable storage detector 4 is fixedly connected to the main control box 1. The image recognizer 6 is slidably connected to the main control box 1 via a guide rail.

[0078] In this embodiment, the edge computing board 124 integrates the functions of the main control unit, and is responsible for all logic control, sensor data acquisition and processing, image recognition, positioning and navigation, communication and task scheduling of the inspection robot. The motor drive board 113, as a lower-level execution module, only responds to motion control commands issued by the edge computing board 124, such as speed, direction, and push rod extension.

[0079] The integrated machine vision and edge computing grain inspection robot also includes a remote monitoring system 5; the remote monitoring system 5 is connected to the main control box 1, and the remote monitoring system 5 is used to collect real-time images in manual remote control mode and send them directly to the remote control center.

[0080] like Figure 2-4 As shown, the main control box 1 is divided into a lower main control box 111, a middle main control box 121 and an upper main control box 131 arranged from top to bottom; the lower main control box 111, the middle main control box 121 and the upper main control box 131 are fixedly connected by screws.

[0081] The lower layer 111 of the main control box is fixedly provided with a battery box 112, a motor drive board 113 and a battery box baffle 114. The battery box 112 is embedded in the bottom of the lower layer 111 of the main control box, and the battery box baffle 114 is connected to the lower layer 111 of the main control box by screws and stops the battery box 112.

[0082] The edge computing board 124, the ambient gas detector 122, and the indoor positioning module 123 are fixedly installed in the middle layer 121 of the main control box; the ambient gas detector 122 is electrically connected to the edge computing board 124 and is used to detect the gas composition in the storage environment and send the detection data to the edge computing board 124.

[0083] The top of the upper layer 131 of the main control box is fixedly connected to the bottom of the remote monitoring system 5.

[0084] Preferably, the indoor positioning module 123 is a UWB-IMU indoor positioning module.

[0085] The bottom of the lower layer 111 of the main control box is provided with a symmetrical longitudinal rod structure, and the rod structure has a long slot along its length. The image recognizer 6 has plug-in parts on both sides. The plug-in parts are nested outside the rod structure and can slide up and down along the rod structure. The height is fixed by tightening the locking screw through the plug-in parts and the long slot to change the field of view height of the camera.

[0086] Preferably, the bottom of the lower layer 111 of the main control box has a supporting and fixing structure, which is nested and connected with the supporting structure 2 to fix the main control box 1.

[0087] In this embodiment, the rod structure is a hollow long slot along its length. Each side of the image recognizer 6 has an elastic connector, which is nested inside the rod structure and can slide up and down along it. When height adjustment is needed, the image recognizer 6 is manually pushed or pulled along the rod structure to the target position, and then the locking screw on the side of the connector is tightened. The screw compresses the connector, causing it to elastically deform and thus grip the rod structure, fixing the height through friction. Loosening the screw allows for readjustment. In this way, the camera's field of view height of the image recognizer 6 can be changed.

[0088] In this embodiment, a hollow structure is located below the lower layer 111 of the main control box for fixing the battery box 112 to the lower layer of the main control box, and also for facilitating the placement and removal of the battery box 112 from the main control box 1. The battery box 112 is electrically connected to all electrical devices, providing all the power for the grain inspection robot integrating machine vision and edge computing of this invention. The battery box 112 is inserted into the lower layer 111 of the main control box in an embedded manner, thereby enabling quick battery replacement during work tasks.

[0089] A motor drive board 113 is fixed above the lower layer 111 of the main control box. The motor drive board 113 serves as an execution-level drive unit and is electrically connected to the edge computing board 124. It is used to receive control commands from the edge computing board 124 and then control the following components: the push rod motor 203 of the support structure 2, the rolling drive motor 307 of the drive system 3, the descending drive motor 403 of the insertion storage detector 4, the rotation servo motor 504 and the rotation servo motor 506 of the remote monitoring system 5, the light source of the remote monitoring system 5, and the light source of the image recognizer 6, thereby realizing the robot's movement, steering, and detection functions.

[0090] The lower layer 111 of the main control box is connected to the battery box baffle 114 by screws, thereby preventing the battery box 112 from falling off.

[0091] The middle layer 121 of the main control box is fixedly connected to the rack and pinion limiter 401 of the insertable storage detector 4, the ambient gas detector 122 and the indoor positioning module 123.

[0092] The ambient gas detector 122 employs an electrochemical sensor, which can effectively detect the concentration and mass of gases such as oxygen, carbon dioxide, nitrogen, and phosphine in the warehouse environment air. The ambient gas detector 122 is electrically connected to the edge computing board 124, transmitting the detected data to the edge computing board 124 for storage and analysis.

[0093] The ambient gas detector 122 employs electrochemical principles and incorporates built-in electrochemical sensing units capable of detecting gases such as oxygen, carbon dioxide, nitrogen, and phosphine. The ambient gas detector 122 is installed in the middle layer of the main control box of the inspection robot, with its air inlet connected to the warehouse environment air. It collects gas samples from the warehouse environment in real time and transmits the detection data to the edge computing board 124. Those skilled in the art can select commercially available electrochemical multi-gas detection modules to achieve the above functions according to actual detection needs.

[0094] In this embodiment, the ambient gas detector 122 includes at least an oxygen sensor, a carbon dioxide sensor, a phosphine sensor, a nitrogen sensor, a VOC sensor, and a temperature and humidity sensor, used to detect the concentrations of the aforementioned gases and temperature and humidity parameters in the warehouse environment air. Exemplarily, the oxygen sensor is a Huashenke 4MZ-HH oxygen sensor, the carbon dioxide sensor is a Huashenke 4MZ-HH-CO2 sensor, the phosphine sensor is a Huashenke 4MZ-HH phosphine sensor, the nitrogen sensor is a Huashenke 4MZ-HH-N2 sensor, the VOC sensor is a Huashenke 4MZ-HH-VOC sensor, and the temperature and humidity sensor is a Huashenke 4MZ-HH temperature and humidity sensor; all are electrochemical sensors. Those skilled in the art will understand that the specific models described above are merely illustrative and not intended to limit the scope of protection of this invention; any electrochemical sensor capable of detecting the corresponding gas concentration can be selected according to actual warehouse needs to achieve the technical effects of this invention.

[0095] The rack and pinion limiter 401 of the insertable storage detector 4 is fixedly connected to the edge computing board 124 by screws.

[0096] The edge computing board 124 is electrically connected to the ambient gas detector 122, the indoor positioning module 123, the temperature and humidity detection probe 404 of the insertable storage detector 4, and the image recognizer 6. It is used to receive sensor data from the ambient gas detector 122, positioning data from the indoor positioning module 123, temperature and humidity data of the storage stack from the insertable storage detector 4, and image information from the image recognizer 6.

[0097] The edge computing board 124 is pre-deployed with a YOLO pest identification model based on the YOLO baseline model. In this embodiment, the YOLO pest identification model is trained using 3000 images of pests on the surface of stored goods and image annotation coordinate files. These 3000 images were taken by a grain depot in Zhenjiang and cover five common stored grain pests: rice weevils, red flour beetles, wheat moths, rusty red flour beetles, and booklice. The annotation files are YOLO format txt files. The training set to test set ratio is 8:2, the model input image size is 640×640, the baseline model is trained for 500 epochs, and the batch size is 16. By comparing the training effects of different versions of the lightweight baseline model, the detection performance of the model is evaluated using precision (P), recall (R), and mean average precision (mAP). The model with the best detection performance is selected as the image recognition model of this invention. In this embodiment, after training, a precision (P) and recall (R) both greater than 85% are used as the screening criteria for effective models. The official YOLO lightweight baseline models that meet the criteria are further evaluated using mean average precision (mAP), and the model with the best detection performance is selected as the image recognition model of this invention. Based on the current dataset, the best-performing version is the YOLO26n model under YOLO26. YOLO26n is an ultra-lightweight (Nano) object detection model in the YOLO26 series, officially open-sourced by Ultralytics in 2026.

[0098] The upper layer 131 of the main control box is fixedly connected to the remote monitoring system 5, thereby facilitating the remote monitoring system 5 to change direction.

[0099] like Figure 5-7 As shown, the support structure 2 includes a driven wheel bracket and a drive wheel bracket, which are located on the left and right sides of the robot and are arranged parallel to each other.

[0100] The driven wheel bracket includes: a cylindrical wheel connecting rod 201, two connecting rods 202, two right-angle connectors 205, and two push rod rotating connectors 207; wherein the two connecting rods 202 are parallel and spaced apart, one end of each connecting rod 202 is connected to one end of the cylindrical wheel connecting rod 201 through a right-angle connector 205, and the other end of each connecting rod 202 is connected to a push rod rotating connector 207;

[0101] The drive wheel bracket includes: another cylindrical wheel connecting rod 201, two push rod motors 203, a support rod 204, two push rod right-angle connectors 206, and two support rod connectors 208; wherein the two push rod motors 203 are arranged in parallel, and the housing of each push rod motor 203 is connected to one end of the other cylindrical wheel connecting rod 201 through the push rod right-angle connector 206; the two ends of the support rod 204 are fixedly connected to the housing of the push rod motor 203 on the same side through the support rod connectors 208; the output shaft of the push rod motor 203 is hinged to the push rod rotating connector 207, and the hinge holes of the two push rod rotating connectors 207 are parallel to each other and located in the same plane.

[0102] One end of the push rod rotating connector 207 is hinged to the output shaft of the push rod motor 203, and the other end is fixedly connected to the connecting rod 202; the housing of the push rod motor 203 is fixed to the drive wheel bracket through the push rod right angle connector 206;

[0103] When turning, only one of the push rod motors operates: when turning left, the push rod of the push rod motor 203 on the right extends outward, while the push rod motor on the left remains stationary. The driven wheel rotates around the hinge point on the left, i.e., the connection point between the left push rod rotating connector 207 and the output shaft of the push rod motor 203, to achieve overall deflection of the driven wheel; the opposite is true when turning right.

[0104] The drive system 3 includes two roller assemblies: a drive wheel and a driven wheel. Each drive wheel and driven wheel includes a lightweight roller 301, an anti-slip layer 302, a roller support 303, a bearing 304, and a bearing fixing member 305. The anti-slip layer 302 is provided on the outer circumference of the lightweight roller 301, and the interior of the lightweight roller 301 is fixedly connected to the roller support 303. The roller support 303 has a central shaft hole, the outer ring of the bearing 304 is fixedly connected to the shaft hole of the roller support 303, and the inner ring of the bearing 304 is fixedly connected to the bearing fixing member 305. The bearing fixing member 305 is fixedly connected to the roller connecting rod 201. The lightweight roller 301 can rotate freely relative to the roller connecting rod 201 via the bearing 304.

[0105] The drive wheel further includes a motor support 306 and a rolling drive motor 307; wherein, the motor support 306 is sleeved on the roller connecting rod 201 in the drive wheel bracket; one end of the motor support 306 is fixedly connected to the housing of the rolling drive motor 307, the output shaft of the rolling drive motor 307 is provided with a drive gear, and the roller support 303 of the drive wheel is provided with a driven gear. The drive gear and the driven gear mesh with each other. When the rolling drive motor 307 rotates, it drives the lightweight roller 301 of the drive wheel to rotate through gear meshing, thereby driving the inspection robot to move as a whole.

[0106] The rolling drive motor 307 and the push rod motor 203 are electrically connected to the motor drive board 113, thereby receiving instructions to work and sending pulse signals to the motor drive board 113, so that the motor drive board 113 can calculate the rotation speed and direction of each motor.

[0107] like Figure 8 As shown, the insertable storage detector 4 includes a rack and pinion limiter 401, a rack 402, a descent drive motor 403, and a temperature and humidity detection probe 404;

[0108] The rack limiting member 401 is a housing structure with a longitudinal sliding groove, and is fixedly installed on the main control box 1; the rack 402 is embedded in the sliding groove of the rack limiting member 401 and can slide up and down along the sliding groove; the housing of the descent drive motor 403 is fixed to the upper end of the rack limiting member 401, and a descent drive gear is installed on the output shaft of the descent drive motor 403; a rack gear structure is provided on one side of the rack 402, and the rack gear structure meshes with the descent drive gear.

[0109] The temperature and humidity detection probe 404 is fixedly connected to the lower end of the rack 402;

[0110] When the descent drive motor 403 rotates forward, it drives the rack 402 to slide downward through gear engagement, causing the temperature and humidity detection probe 404 to insert into the surface of the storage stack. When the descent drive motor 403 rotates in reverse, it drives the rack 402 to slide upward, causing the temperature and humidity detection probe 404 to leave the surface of the storage stack. The insertion depth can be preset according to the thickness of the storage stack. Because the diameter of the temperature and humidity detection probe 404 is small, its insertion has no or negligible impact on the flatness of the storage surface.

[0111] The motor drive board 113 is electrically connected to the descent drive motor 403 and is used to monitor the power of the descent drive motor 403 and send the power value to the edge computing board 124 in real time. The edge computing board 124 is used to determine whether the power value exceeds a preset threshold, identify the clumping or caking of the surface of the storage stack according to the determination result, and trigger an abnormal warning when the threshold is exceeded.

[0112] In this embodiment, when the descent drive motor 403 receives the instruction from the motor drive board 113 to rotate, it can drive the rack 402 along with the temperature and humidity detection probe 404 to descend or rise, so that the temperature and humidity detection probe 404 is inserted into the surface of the storage. The insertion depth can be preset according to the thickness of the storage stack, generally 0-60cm. Since the diameter of the temperature and humidity detection probe 404 is small, when the temperature and humidity detection probe 404 is inserted into the surface of the storage, it has no effect on the flatness of the storage surface or the effect is negligible. Meanwhile, the motor drive board 113 monitors the power of the descent drive motor 403 (the power of the descent drive motor 403 can reflect the resistance when the insertion storage detector 4 is inserted into the storage stack, and the change in resistance can be used to further determine the clumping or hardening of the storage stack surface), and sends the power value to the edge computing board 124 in real time. The edge computing board 124 determines the clumping or hardening of the storage stack surface based on whether the power value exceeds a preset threshold: when there is no abnormality in the storage, the descent drive motor 403 maintains a stable and low power; when the storage exhibits abnormalities such as hardening or clumping, the probe contacts the abnormal area, causing a significant jump in power that continues to increase. Based on the preset power threshold, when the power of the descent motor 403 exceeds the threshold, an abnormality warning is triggered.

[0113] In this embodiment, the calibration method for the power threshold of the drive motor 403 is as follows:

[0114] First, multiple insertion tests are conducted on a normal storage surface to collect power data of the descent drive motor 403 under normal conditions, and its average power value is calculated as the normal baseline. Second, abnormal storage samples with agglomeration and compaction are artificially created, such as by compaction or moisture absorption to induce local compaction of the storage (artificially created abnormal samples can simulate the resistance characteristics of natural agglomeration / compaction, and those skilled in the art can adjust them according to common agglomeration forms in actual storage). Multiple insertion tests are conducted in the same storage environment to collect power data of the descent drive motor 403 under abnormal conditions, and the power variation range under abnormal conditions is obtained. Finally, by comparing the normal baseline data with the abnormal power range, a warning power threshold is set based on actual detection requirements.

[0115] This threshold can be calibrated according to different storage types (such as rice, wheat, corn, etc.) and different degrees of compaction. It should be noted that the required insertion pressure varies for different storage types and different degrees of compaction. The power threshold of this invention does not need to be calculated using a uniform formula. Those skilled in the art can obtain the applicable threshold parameters through experimental measurement for specific storage environments by following the above calibration method.

[0116] like Figure 9-10 As shown, the remote monitoring system 5 includes a camera mounting component 501, a camera upper cover 502, a camera lower cover 503, a first rotating servo motor 504, a camera rotating component 505, a second rotating servo motor 506, and a first camera and a first light source integrated therein.

[0117] The upper camera cover 502, lower camera cover 503, and rotating servo motor 504 are all fixedly connected to the camera mounting bracket 501 by screws. The camera mounting bracket 501 has two slots, which are used to fix the first camera and the first light source, respectively. The rotating servo motor 504 and the rotating servo motor 506 are fixedly connected to the camera rotating component 505, and both are electrically connected to the motor drive board 113 to receive commands to rotate and change the image acquisition angle of the first camera.

[0118] The remote monitoring system 5 is fixed to the top of the upper layer 131 of the main control box. The first camera is used to collect real-time images in manual remote control mode. The image data is sent directly to the remote control center for the operator to view, without being stored or processed by the edge computing board 124. The first light source is electrically connected to the motor drive board 113 and can be remotely turned on by the operator to provide lighting when the light is insufficient.

[0119] The image recognizer 6 integrates a second camera and a second light source, capable of capturing images in a bright field of view. The second camera is used for pest identification in automatic inspection mode, and the image data it captures is transmitted to the edge computing board 124, where it is detected and analyzed by the YOLO pest identification model pre-deployed in the edge computing board 124. The second light source is automatically activated by the edge computing board 124 during automatic inspection to ensure image quality.

[0120] As shown in Tables 1-3, through further comparison of the training results of three preferred lightweight models (YOLOv8n, YOLO11n, and YOLO26n), and based on precision, recall, mAP50, and the actual test results, YOLO26n was further selected as the baseline model for this invention. After 400 training rounds, the precision of this model tended to stabilize and was higher than 0.90, and the recall also tended to stabilize and was higher than 0.92. When this model was used for warehouse inspection, it could effectively detect the types and quantities of pests on the surface of rice and wheat stored goods, with generally high accuracy, which can meet the needs of this invention for pest image detection.

[0121] Table 1 Performance of YOLOv8n Model

[0122]

[0123] Table 2 Performance of YOLO11n Model

[0124]

[0125] Table 3 Performance of the YOLO26n model

[0126]

[0127] Figure 11-14 These are all field test results validation images of the YOLO26n lightweight detection model for stored pests in rice and wheat. They are used to verify the model's detection accuracy, recognition stability, and applicability to real-world storage conditions. The results are cross-validated with the precision, recall, and mAP50 data in Tables 1–3, as detailed below:

[0128] Figure 11 The results of basic detection of storage pests by the YOLO26n model on a single sample are shown. The confidence scores for the detection of rice weevil (Oryzae) are 0.84 and for the detection of Tribolium (Tribolium) are 0.38. This verifies the model's basic ability to identify typical storage pests and demonstrates that the initial detection confidence score for the major pest, rice weevil, meets the standard.

[0129] Figure 12 The results of batch detection of multiple pests by the YOLO26n model are shown. The confidence level for rice weevil detection is 0.81, and the confidence levels for pseudo-floribunda detection are 0.87 / 0.88 / 0.84, respectively. This proves that the model has an overall confidence level of over 0.84 for identifying target pests in the scenario of multiple pests coexisting, and the multi-target detection accuracy is stable and reliable.

[0130] Figure 13The YOLO26n model is demonstrated to perform well in mixed storage of rice and wheat and mixed pest samples. The confidence scores for rice weevil detection are 0.80 / 0.84 and for false grain beetle detection are 0.85. This verifies that the model consistently achieves a confidence score of over 0.80 for pest identification in complex storage scenarios, and the detection accuracy in mixed scenarios meets the standards.

[0131] Figure 14 The results of pest detection using the YOLO26n model in a real warehouse environment are shown. The confidence scores for rice weevil detection are 0.68 / 0.28 / 0.77, and the confidence score for pseudo-grass beetle detection is 0.78. This demonstrates that the model can still effectively identify pests under complex conditions such as lighting and debris interference in a real warehouse, and has the robustness and practicality for real-world applications.

[0132] Figure 11-14 The results demonstrate that the YOLO26n model can accurately identify the types and quantities of pests on the surface of stored rice and wheat, and the detection confidence level meets the requirements of actual storage. Combined with the precision, recall, and mAP50 indices in Table 1-3, the model is found to be stable in training and accurate enough to meet the storage pest image detection requirements of this invention.

[0133] Specifically, the working principle of this invention is as follows:

[0134] Working Mode 1: The edge computing board 124 of the inspection robot can communicate with the remote control center. Operators can control the inspection robot by sending control commands. After receiving the command, the edge computing board 124 initializes the inspection robot and then parses and controls it according to the control command. Simultaneously, the first camera of the remote monitoring system 5 provides the operator with a working field of view by acquiring real-time images, allowing them to observe the inspection robot's working status and understand the environmental conditions. If the warehouse is dimly lit, the operator can also send a command to the motor drive board 113 to control the first light source switch of the remote monitoring system 5, thereby sending a brighter image. The operator can also send commands to the edge computing board 124 to control the rotation of servo motors 504 and 506, thereby changing the image viewing angle of the first camera of the remote monitoring system 5 to monitor the inspection robot's working status and the status of the stored goods. It should be noted that the second camera of the image recognizer 6 in manual mode is only controlled by the edge computing board 124 to acquire images and perform recognition when the operator manually triggers the pest detection command. Meanwhile, the edge computing board 124 continuously communicates with the indoor positioning module 123 to continuously obtain the current position and orientation of the inspection robot, and continuously sends it back to the remote control center for convenient and flexible control by the operator.

[0135] When a movement command is received, the edge computing board 124 sends a drive command to the motor drive board 113. After parsing the command, the motor drive board 113 controls the rolling drive motor 307 of the drive system 3 to drive the drive wheel to rotate, thereby moving the inspection robot as a whole. At the same time, the motor drive board 113 controls the extension and retraction of the push rod motor 203 to adjust the rotation angle of the driven wheel, thereby changing the forward direction of the inspection robot and realizing the movement task of the inspection robot.

[0136] Once the operator controls the inspection robot to reach the inspection area, they can send a detection command. Upon receiving the command, the edge computing board 124 communicates with the environmental gas detector 122 to collect the air quality and component concentration at the current task point, and stores and sends the detection results to the remote control center. Simultaneously, the operator can send a command to control the descent drive motor 403 via the motor drive board 113, causing the insertion storage detector 4 to insert into the surface of the storage stack. The motor drive board 113 monitors the power of the descent drive motor 403 to determine the degree of clumping or hardening on the storage stack surface. Meanwhile, the edge computing board 124 communicates with the temperature and humidity detection probe 404 to obtain the temperature and humidity information of the storage stack at the current task point, stores the detection results, and sends them back to the remote monitoring center. Subsequently, the operator continues to control the descent drive motor 403 to retract, causing the insertion storage detector 4 to leave the storage stack surface. Meanwhile, operators can communicate with the image recognizer 6 via the edge computing board 124 by sending commands to acquire images of the warehouse stack surface. After receiving the images, the edge computing board 124 identifies the acquired images using a pre-deployed pest identification model, records the types and quantities of pests identified, and sends the data to a remote monitoring center. If the warehouse is dimly lit, operators can send commands to control the motor drive board 113 to turn on the light source of the image recognizer 6, thereby acquiring images of the warehouse stack in bright light conditions.

[0137] After the inspection work at the previous task point is completed, the operator can continue to send commands to guide the inspection robot to move to the next task point and perform the above inspection tasks.

[0138] Once all task points have been inspected, the operator can send commands to move the inspection robot to a standby point for the next task. Simultaneously, the edge computing board 124 organizes and packages all information collected by the inspection robot for local storage and sends it to cloud storage.

[0139] Subsequently, the operator sends a command to put the inspection robot into a low-power standby mode, completing one manual inspection task.

[0140] Working Mode 2: The operator sends an automatic inspection command to the inspection robot to perform automatic inspection tasks. After receiving the command, the edge computing board 124 initializes and calibrates the robot as a whole. After calibration, the edge computing board 124 continuously communicates with the indoor positioning module 123 to obtain the robot's position and orientation in real time. After obtaining its own positioning, the edge computing board 124 confirms the target working point location through the preset task point. At the same time, it sends real-time drive commands to the motor drive board 113, which controls the rolling drive motor 307 of the drive system 3 and the push rod motor 203 of the support structure 2, thereby enabling the inspection robot to move on the warehouse stack surface. With the assistance of the indoor positioning module 123, the edge computing board 124 continuously adjusts the drive commands, enabling the inspection robot to move autonomously to the task point. The entire automatic inspection process does not use the remote monitoring system 5, but only relies on the image recognition device 6 for visual acquisition.

[0141] When the inspection robot arrives at the task point, the edge computing board 124 communicates with the ambient gas detector 122 to obtain and record the ambient gas quality and composition information at the current task point. Simultaneously, the edge computing board 124 sends a drive command to the motor drive board 113 to control the rotation of the descent drive motor 403, thereby causing the insertable storage detector 4 to insert into the lower stack. The motor drive board 113 monitors the power of the descent drive motor 403 to determine the surface caking or compaction of the storage stack. The edge computing board 124 communicates with the temperature and humidity detection probe 404 to obtain and record the temperature and humidity information of the current task point stack and whether caking or compaction has occurred. Simultaneously, the edge computing board 124 sends a command to turn on the second light source of the image recognizer 6 via the motor drive board 113. The edge computing board 124 communicates with the second camera of the image recognizer 6 to obtain a bright image of the warehouse stack surface. The edge computing board 124 uses a pre-deployed YOLO pest identification model to identify the image of the warehouse stack surface, thereby determining and recording the type and quantity of pests at the current task point. When any of the above indicators exceeds a set threshold, the edge computing board 124 records the task point and sends an early warning message and the current location coordinates to the remote monitoring center for easy handling by operators. If there are no abnormalities, the edge computing board 124 sends a command to turn off the second light source of the image recognizer 6 and retrieves the insertable storage detector 4, removing it from the stack. The inspection task for this task point is completed.

[0142] After the previous task point inspection is completed, the edge computing board 124 confirms the next task point and controls the support architecture 2 and drive system 3 again, and repeats the above task until all task points are inspected.

[0143] After all task points have been inspected, the edge computing board 124 confirms the coordinates of the standby point and then sends a command to control the inspection robot to return to the standby point autonomously.

[0144] After the inspection robot returns to the standby point, the edge computing board 124 organizes, packages, stores, and uploads all the information collected during this inspection task to the cloud.

[0145] The inspection robot enters low-power sleep mode. It completes one automated inspection task.

[0146] In summary, the automatic inspection method of the grain inspection robot integrating machine vision and edge computing of the present invention includes the following steps:

[0147] Step S1: The remote operation center sends an automatic inspection command, and the edge computing board 124 initializes and calibrates the robot after receiving the command;

[0148] Step S2: The edge computing board 124 obtains the robot's position and orientation in real time through the indoor positioning module 123, and controls the drive system 3 to make the robot move autonomously to the preset task point;

[0149] Step S3: After reaching the task point, the edge computing board 124 controls the ambient gas detector 122 to collect ambient gas data, controls the insertable storage detector 4 to insert into the storage stack to collect temperature and humidity data and monitors the power of the descent drive motor 403 to determine the clumping situation, and controls the image recognizer 6 to collect images of the storage stack surface.

[0150] Step S4: Edge computing board 124 runs the pre-deployed YOLO pest identification model to identify the collected images and record the types and quantities of pests;

[0151] Step S5: When any collected data exceeds the set threshold, the edge computing board 124 sends an early warning message and its current location to the remote monitoring center;

[0152] Step S6: After completing the current task point, the robot moves to the next task point and repeats steps S3 to S5 until all task points have been inspected.

[0153] Step S7: The robot returns to the preset standby point, stores all detection data locally, and uploads it to the cloud.

[0154] This invention employs an autonomous, mobile inspection robot structure, capable of covering different areas of large warehouses. It overcomes the spatial limitations of traditional fixed single-point monitoring, enabling the acquisition of environmental parameter anomalies caused by spatial heterogeneity in different areas, thus reducing monitoring blind spots. The integrated insertable storage detector 4 can directly penetrate the warehouse stacks to acquire internal temperature and humidity data. Combined with the environmental gas detector 122, it can simultaneously acquire multi-dimensional environmental information, providing more comprehensive references compared to traditional surface monitoring data. Through the edge computing board 124, image recognition and analysis of pest anomalies on the warehouse stack surface can be performed directly on the robot, eliminating the need to upload large amounts of raw image data to the cloud. This reduces network bandwidth pressure, shortens the data transmission link for anomaly recognition and response, and enables real-time monitoring and early warning. This invention can replace the traditional manual inspection mode, reducing manpower investment and long-term maintenance costs. It enables automated inspection around the clock, providing automation and data traceability capabilities for warehouse environmental monitoring. It can adapt to the differentiated monitoring needs of different types of stored materials, offering considerable application flexibility.

[0155] It should be understood that although this specification is described according to various embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

[0156] The detailed descriptions listed above are merely specific illustrations of feasible embodiments of the present invention and are not intended to limit the scope of protection of the present invention. All equivalent embodiments or modifications made without departing from the spirit of the present invention should be included within the scope of protection of the present invention.

Claims

1. A grain inspection robot integrating machine vision and edge computing, characterized in that, It includes a main control box (1), a support structure (2), a drive system (3), an insertable storage detector (4), an image recognizer (6), an edge computing board (124), an indoor positioning module (123), and an ambient gas detector (122). The main control box (1) is fixed on the support structure (2), the drive system (3) is connected to the support structure (2) and supports the support structure (2), and the drive system (3) is used to drive the robot to move and turn; The insertable storage detector (4) is installed on the main control box (1) and is used to insert into the storage stack to collect temperature and humidity data; The image recognizer (6) is slidably connected to the bottom of the main control box (1) and is used to collect images of the surface of the warehouse stacks; The indoor positioning module (123) is installed inside the main control box (1) and electrically connected to the edge computing board (124) to obtain the robot's position and orientation; The ambient gas detector (122) is installed inside the main control box (1) and electrically connected to the edge computing board (124) to detect the gas composition in the storage environment; The edge computing board (124) is located in the main control box (1) and serves as the main control unit. It is electrically connected to the indoor positioning module (123), the environmental gas detector (122), the drive system (3), the insertable storage detector (4), and the image recognition device (6), respectively. It is used to receive remote instructions, process the data of the insertable storage detector (4), recognize the image of the image recognition device (6), plan the inspection path, and send drive instructions to the drive system (3).

2. The grain inspection robot integrating machine vision and edge computing according to claim 1, characterized in that, It also includes a remote monitoring system (5); The remote monitoring system (5) is connected to the main control box (1). The remote monitoring system (5) is used to collect real-time images in manual remote control mode and send them directly to the remote control center.

3. The grain inspection robot integrating machine vision and edge computing according to claim 1, characterized in that, The support structure (2) includes a driven wheel bracket and a drive wheel bracket, which are located on the left and right sides of the robot and are arranged parallel to each other. The driven wheel bracket includes: a cylinder wheel connecting rod (201), two connecting rods (202), two right-angle connectors (205), and two push rod rotating connectors (207); wherein the two connecting rods (202) are parallel and spaced apart, one end of each connecting rod (202) is connected to one end of the cylinder wheel connecting rod (201) through a right-angle connector (205), and the other end of each connecting rod (202) is connected to a push rod rotating connector (207); The drive wheel bracket includes: another cylindrical wheel connecting rod (201), two push rod motors (203), a support rod (204), two push rod right-angle connectors (206), and two support rod connectors (208); wherein the two push rod motors (203) are arranged in parallel, and the housing of each push rod motor (203) is connected to one end of the other cylindrical wheel connecting rod (201) through the push rod right-angle connector (206); the two ends of the support rod (204) are fixedly connected to the housing of the push rod motor (203) on the same side through the support rod connector (208); the output shaft of the push rod motor (203) is hinged to the push rod rotating connector (207), and the hinge holes of the two push rod rotating connectors (207) are parallel to each other and located in the same plane.

4. The grain inspection robot integrating machine vision and edge computing according to claim 3, characterized in that, One end of the push rod rotating connector (207) is hinged to the output shaft of the push rod motor (203), and the other end is fixedly connected to the connecting rod (202); the housing of the push rod motor (203) is fixed to the drive wheel bracket through the push rod right angle connector (206); When turning, only one of the push rod motors operates: when turning left, the push rod of the push rod motor (203) on the right extends outward, while the push rod motor on the left remains stationary. The driven wheel rotates around the hinge point on the left, i.e., the connection point between the left push rod rotating connector (207) and the output shaft of the push rod motor (203), to achieve overall deflection of the driven wheel; the opposite is true when turning right.

5. The grain inspection robot integrating machine vision and edge computing according to claim 3, characterized in that, The drive system (3) includes two roller assemblies, namely a drive wheel and a driven wheel; Each of the driving wheel and the driven wheel includes: a lightweight roller wheel (301), an anti-slip layer (302), a roller wheel support (303), a bearing (304), and a bearing fixing member (305); wherein, the anti-slip layer (302) is provided on the outer circumference of the lightweight roller wheel (301), and the interior of the lightweight roller wheel (301) is fixedly connected to the roller wheel support (303); the center of the roller wheel support (303) is provided with a shaft hole, the outer ring of the bearing (304) is fixedly connected to the shaft hole of the roller wheel support (303), and the inner ring of the bearing (304) is fixedly connected to the bearing fixing member (305); the bearing fixing member (305) is fixedly connected to the roller wheel connecting rod (201); the lightweight roller wheel (301) can rotate freely relative to the roller wheel connecting rod (201) through the bearing (304); The drive wheel further includes: a motor support (306) and a rolling drive motor (307); wherein, the motor support (306) is sleeved on the roller connecting rod (201) in the drive wheel bracket; one end of the motor support (306) is fixedly connected to the housing of the rolling drive motor (307), the output shaft of the rolling drive motor (307) is provided with a drive gear, the roller support (303) of the drive wheel is provided with a driven gear, the drive gear and the driven gear mesh with each other, and when the rolling drive motor (307) rotates, the lightweight roller (301) of the drive wheel is driven to rotate through the gear meshing.

6. The grain inspection robot integrating machine vision and edge computing according to claim 1, characterized in that, The insertable storage detector (4) includes a rack limiter (401), a rack (402), a descent drive motor (403), and a temperature and humidity detection probe (404). The rack limiting member (401) is a housing structure with a longitudinal sliding groove, and is fixedly installed on the main control box (1); the rack (402) is embedded in the sliding groove of the rack limiting member (401) and can slide up and down along the sliding groove; the housing of the descent drive motor (403) is fixed to the upper end of the rack limiting member (401), and a descent drive gear is installed on the output shaft of the descent drive motor (403); a rack gear structure is provided on one side of the rack (402), and the rack gear structure meshes with the descent drive gear; The temperature and humidity detection probe (404) is fixedly connected to the lower end of the rack (402); When the descent drive motor (403) rotates forward, it drives the rack (402) to slide downward through gear meshing, so that the temperature and humidity detection probe (404) is inserted into the storage stack surface; when the descent drive motor (403) rotates in reverse, it drives the rack (402) to slide upward, so that the temperature and humidity detection probe (404) leaves the storage stack surface.

7. The grain inspection robot integrating machine vision and edge computing according to claim 6, characterized in that, The main control box (1) is also equipped with a motor drive board (113), which is electrically connected to the descent drive motor (403) and is used to monitor the power of the descent drive motor (403) and send the power value to the edge computing board (124) in real time. The edge computing board (124) is used to determine whether the power value exceeds a preset threshold, identify the clumping or caking of the surface of the storage stack according to the determination result, and trigger an abnormal warning when the threshold is exceeded.

8. The grain inspection robot integrating machine vision and edge computing according to claim 7, characterized in that, The main control box (1) is divided into a lower layer (111), a middle layer (121), and an upper layer (131) arranged from top to bottom. The lower layer (111) of the main control box is fixedly provided with a battery box (112), a motor drive board (113) and a battery box baffle (114). The battery box (112) is embedded in the bottom of the lower layer (111) of the main control box. The battery box baffle (114) is connected to the lower layer (111) of the main control box by screws and stops the battery box (112). The edge computing board (124), the ambient gas detector (122), and the indoor positioning module (123) are fixedly installed in the middle layer (121) of the main control box; the ambient gas detector (122) is electrically connected to the edge computing board (124) and is used to detect the gas composition in the storage environment and send the detection data to the edge computing board (124). The top of the upper layer (131) of the main control box is fixedly connected to the bottom of the remote monitoring system (5).

9. The grain inspection robot integrating machine vision and edge computing according to claim 8, characterized in that, The bottom of the lower layer (111) of the main control box is provided with a symmetrical longitudinal rod structure, and the rod structure has a long slot along its length. The image recognizer (6) is provided with plug-in parts on both sides. The plug-in parts are nested outside the rod structure and can slide up and down along the rod structure. The height is fixed by tightening the locking screw through the plug-in parts and the long slot to change the field of view height of the camera.

10. An inspection method for a grain inspection robot based on any one of claims 1 to 9, integrating machine vision and edge computing, characterized in that, Includes the following steps: Step S1: The remote operation center sends an automatic inspection command, and the edge computing board (124) initializes and calibrates the robot after receiving the command; Step S2: The edge computing board (124) obtains the robot's position and orientation in real time through the indoor positioning module (123) and controls the drive system (3) to enable the robot to move autonomously to the preset task point; Step S3: After reaching the task point, the edge computing board (124) controls the ambient gas detector (122) to collect ambient gas data, controls the insertable storage detector (4) to insert into the storage stack to collect temperature and humidity data and monitor the power of the descent drive motor (403) to determine the clumping situation, and controls the image recognition device (6) to collect images of the storage stack surface. Step S4: The edge computing board (124) runs the pre-deployed YOLO pest identification model to identify the collected images and record the types and quantities of pests; Step S5: When any collected data exceeds the set threshold, the edge computing board (124) sends an early warning message and its current location to the remote monitoring center; Step S6: After completing the current task point, the robot moves to the next task point and repeats steps S3 to S5 until all task points have been inspected. Step S7: The robot returns to the preset standby point, stores all detection data locally, and uploads it to the cloud.

Images