A workpiece falling detection system and method based on the inside of a high-temperature furnace body

Abstract

The application discloses a workpiece falling detection system and method based on the inside of a high-temperature furnace body. The application aims to solve the technical problems of high missing rate of manual inspection, short service life of contact sensors, poor anti-interference ability of traditional visual detection, and difficult maintenance in the existing workpiece falling detection of high-temperature furnace body. The system opens a cavity through the upper and lower sides of the furnace body, installs a multi-layer composite sealed window glass at the inside opening of the cavity, and installs a camera and a light source at the outside opening of the cavity, thereby forming an external detection architecture and a dark field illumination detection mode. The side wall of the furnace body facing the camera adopts a corrugated plate structure, and cooperates with a bending type chassis with grating holes to realize efficient heat dissipation and stray light shielding. The corresponding detection method realizes accurate identification of workpiece falling by establishing an adaptive reference background model, combining multi-dimensional judgment of area, continuous frame number and motion direction.

Description

Technical Field

[0001] This invention relates to the field of automatic detection technology, specifically to a workpiece drop detection system and method based on the interior of a high-temperature furnace. Background Technology

[0002] In industrial production processes such as metal heat treatment, powder metallurgy sintering, and ceramic firing, high-temperature furnaces are core production equipment. Workpieces are typically continuously transported within the furnace via conveyor belts to complete heating, holding, and cooling processes. During production, workpieces frequently fall to the furnace bottom due to factors such as workpiece placement deviations, conveyor belt vibrations, and airflow disturbances within the furnace. If these fallen workpieces are not detected and removed promptly, they can accumulate at the furnace bottom, clogging the furnace cavity, scratching the conveyor belt, and even causing serious equipment malfunctions such as furnace deformation and damage to heating elements. This can lead to unplanned production line downtime and significant economic losses.

[0003] Currently, in the detection of workpiece drops in high-temperature furnaces, the furnace body is usually a closed structure with dim lighting and extremely high temperatures. Manual inspection can only be conducted intermittently through observation holes, making 24-hour real-time monitoring impossible. This easily leads to missed detection of dropped workpieces and poses a safety hazard of burns. Furthermore, traditional contact sensors (such as limit switches and pressure sensors) need to be installed at the bottom of the furnace, directly exposed to the high-temperature environment, making them highly susceptible to oxidation and corrosion, resulting in a very short average lifespan and extremely high maintenance costs. Simultaneously, existing bright-field illumination visual inspection systems are severely affected by furnace heat radiation, dust, and smoke, resulting in extremely low contrast between the workpiece and the background, leading to reduced recognition accuracy and a false alarm rate exceeding 15%, failing to meet industrial production needs. In addition, while traditional solutions that integrate detection equipment into the furnace offer better detection results, equipment replacement and maintenance require furnace shutdown and cooling, resulting in long maintenance times and severely impacting production continuity.

[0004] In summary, existing high-temperature furnace workpiece drop detection technologies cannot simultaneously solve the problems of real-time performance, accuracy, reliability, and ease of maintenance, making them unsuitable for the requirements of large-scale continuous production in industrial heat treatment. There is an urgent need to develop a non-contact, interference-resistant, long-life, and easy-to-maintain high-temperature furnace workpiece drop detection system and method. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a non-contact, anti-interference, high-accuracy, long-life, and easy-to-maintain workpiece drop detection system and method based on the interior of a high-temperature furnace. To achieve the above objective, this invention adopts the following technical solution:

[0006] The first aspect of this invention provides a workpiece drop detection system based on the interior of a high-temperature furnace, comprising:

[0007] Furnace body 1;

[0008] The upper cavity 2 is recessed and formed in the upper part of the furnace body 1;

[0009] The lower cavity 3 is recessed and formed in the lower part of the furnace body 1;

[0010] The upper viewing window 4 is installed over the opening of the upper cavity 2;

[0011] The lower viewing window 5 is installed over the opening of the lower cavity 3;

[0012] Camera 6 is mounted on the outside of the upper cavity 2 via the first base frame 71 and is used to detect workpieces falling into the furnace through the upper viewing window 4.

[0013] The light source 8 is installed on the outside of the lower cavity 3 via the second base frame 72, and is used to provide light to the furnace through the lower viewing window glass 5.

[0014] Preferably, the upper viewing window 4 adopts a multi-layer composite sealing structure, including an inner pressure plate 41, a first silicone pad 42, a high-temperature tempered glass 43, a second silicone pad 44, and an outer pressure plate 45 stacked sequentially from the inside to the outside; wherein, the inner pressure plate 41 and the outer pressure plate 45 press the high-temperature tempered glass 43 into the opening of the upper cavity 2 of the furnace body by bolts 46; the lower viewing window 5 has the same structure as the upper viewing window 4.

[0015] Preferably, the first base frame 71 includes a first planar plate 711 extending into the upper cavity 2 and a second planar plate 712 formed by bending the first planar plate 711 downward and extending outward from the upper cavity 2. The first planar plate 711 is evenly provided with a plurality of first grid holes 713, and the second planar plate 712 is used for mounting the camera 6 thereon.

[0016] The second base frame 72 includes a third planar plate 721 extending into the lower cavity 3, and a fourth planar plate 722 formed by bending the third planar plate 721 downward and extending it outward from the lower cavity 3. The third planar plate 721 is evenly provided with a plurality of second grid holes 723, and the fourth planar plate 722 is used for mounting the light source 8.

[0017] Preferably, the camera 6 is mounted on the first base frame 71 via an adjusting bracket 61; the light source 8 is a spotlight, mounted on the second base frame 72 via an adjusting support 81; the adjusting bracket 61 is used to adjust the pitch angle of the camera 6, and the adjusting support 81 is used to adjust the pitch angle of the light source 7, so that the optical axes of the two can be flexibly aligned with the falling workpiece; wherein, the camera 6 faces the upper cavity 2 and is tilted downward via the adjusting bracket 61, and the light source 8 faces the lower cavity 3 and is tilted upward via the adjusting support 81.

[0018] Preferably, the side wall of the furnace body 1 facing the camera 6 is a corrugated plate 101, and the other side walls are corrugated plates or flat steel plates.

[0019] A second aspect of the present invention provides a method for detecting workpiece drop inside a high-temperature furnace, characterized in that it is applied to the workpiece drop detection system described in the first aspect, and includes the following steps:

[0020] Step S1: Establish a baseline background model when no workpieces fall into the furnace;

[0021] Step S2: Under dark lighting conditions, acquire a continuous image sequence of the detection area inside the furnace;

[0022] Step S3: Compare each frame of the continuous image sequence with the reference background model in turn to extract candidate targets;

[0023] Step S4: Determine whether a workpiece has fallen based on the area, number of consecutive frames, and direction of motion of the candidate target.

[0024] Step S5: If it is determined that the workpiece has fallen, output the falling signal and / or the counting signal.

[0025] Preferably, step S1 includes:

[0026] Step S11: Power on the camera (6) and the light source (8);

[0027] Step S12: Under the condition that no workpieces fall, continuously acquire the initial background image inside the furnace for a preset number of frames;

[0028] Step S13: Perform multi-dimensional denoising and dark field feature filtering on each frame of the initial background image in sequence;

[0029] Step S14: The preprocessed initial background image is fused at the pixel level using the multi-frame averaging method to obtain the initial reference background image B0(x,y), where (x,y) are the image pixel coordinates; at the same time, the gray standard deviation σ(x,y) of each pixel in the multi-frame preprocessed image is calculated and used as the background fluctuation tolerance threshold of the pixel, which is used to distinguish the background fluctuation from the real target in subsequent steps; the initial reference background image B0(x,y) and the gray standard deviation σ(x,y) data are stored to construct the reference background model.

[0030] Preferably, step S13 includes:

[0031] Median filtering is used to remove salt-and-pepper noise generated by thermal radiation inside the furnace and transient interference from suspended dust particles;

[0032] Gaussian filtering is used to smooth local grayscale fluctuations caused by thermal stress deformation of the window glass;

[0033] Based on the principle of low grayscale background and high grayscale of fallen workpiece under dark lighting, an adaptive grayscale threshold T1 is set. Pixels with grayscale values ​​lower than the grayscale threshold T1 are retained as valid background pixels, while isolated pixels with grayscale values ​​higher than the grayscale threshold T1 and connected region area smaller than the preset minimum noise area S1 are identified as thermal noise points and removed.

[0034] Preferably, step S1 further includes:

[0035] Step S15: During the detection process, if no workpiece falls within the preset time limit, the background model update mechanism is activated, and steps S12 and S13 are executed; then, the weighted average method is used to slowly update the reference background model, and the update formula is:

[0036] B k (x,y)=α·B k₋1 (x,y)+(1-α)·B n (x,y);

[0037] Among them, B k (x,y) represents the updated k-th generation baseline background image, B k₋1 (x,y) represents the (k-1)th generation baseline background image before the update, B n (x,y) represents the average image of the n preprocessed frames collected in this acquisition, and α is the preset background update weight;

[0038] Step S16: If a candidate target is found in the pre-detection before background update, or if the overall gray value change of the window glass area exceeds the preset threshold T2, the current background update operation is immediately abandoned, and a window cleaning prompt signal is triggered. The background acquisition and update process is then re-executed after the anomaly is resolved.

[0039] Preferably, step S2 includes: a light source (8) tilting upwards to illuminate the area inside the furnace to form dark field illumination; and a camera (6) tilting downwards to acquire a continuous image sequence I_t(x,y) in real time, where t represents the frame number;

[0040] Preferably, step S3 involves sequentially comparing each frame of the image sequence with a reference background model to extract candidate targets, including:

[0041] Step S31: Perform pixel-by-pixel difference comparison between the current frame image I_t(x,y) and the reference background image B(x,y), and calculate the difference image D_t(x,y)=|I_t(x,y)-B(x,y)|;

[0042] Step S32: Based on the grayscale standard deviation σ(x,y) of each pixel in the baseline background model, set the adaptive difference threshold T_diff(x,y)=k·σ(x,y), where k is a preset scaling factor;

[0043] Step S33: Binarize the difference image and mark the pixels where D_t(x,y)>T_diff(x,y) as candidate target pixels to obtain the binarized image F_t(x,y);

[0044] Step S34: Perform morphological opening operation on the binarized image F_t(x,y), first eroding and then dilating to remove isolated noise points and small interference regions;

[0045] Step S35: Perform connected component labeling on the binarized image after morphological processing, and extract the area, centroid position, and circumscribed rectangle features of each candidate target region;

[0046] Preferably, step S4, determining whether a workpiece has fallen based on the area, number of consecutive frames, and direction of motion of the candidate target, includes:

[0047] Step S41: Calculate the area A_i of each candidate target region. When the preset lower limit of the workpiece area ≤ A_i ≤ the preset upper limit of the workpiece area, it is retained as a valid candidate target.

[0048] Step S42: Perform target association tracking on candidate targets in a continuous frame sequence and record the number of consecutive frames C in which the same target appears;

[0049] Step S43: When C ≥ the preset threshold for the number of consecutive frames, proceed to the next step of judgment;

[0050] Step S44: Calculate the centroid displacement vector of the candidate target in consecutive frames and determine the angle θ between its motion direction and the direction of gravity;

[0051] Step S45: When θ ≤ the preset maximum allowable deflection angle, determine the candidate target as a workpiece falling event;

[0052] Preferably, step S5, if it is determined that a workpiece has fallen, outputs a falling signal and / or a counting signal, including:

[0053] Step S51: When a workpiece falling event is determined, immediately output a falling alarm signal;

[0054] Step S52: Simultaneously update the workpiece drop count value and output the workpiece count signal;

[0055] Step S53: Save the image data at the moment of the fall, including the current frame image and several frames before and after, for post-event retrospective analysis;

[0056] Step S54: If no new falling event is detected within the preset time window, the detection state is automatically reset and the system waits for the next detection.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] 1. This invention's detection system utilizes a layered, externally mounted dark-field illumination detection architecture to achieve non-contact, real-time detection of workpieces falling inside high-temperature furnaces. This solves the problems of existing contact sensors being easily damaged by high temperatures and the difficulty of maintaining built-in equipment. The overall system lifespan is extended from less than one month to over 18 months, significantly reducing maintenance costs. Simultaneously, the dark-field illumination technology enhances the contrast between the workpiece and the background, improving detection accuracy and reducing false alarm rates, thus addressing the poor anti-interference capabilities of traditional visual inspection systems.

[0059] 2. The detection system of this invention achieves both high sealing performance and high thermal stability through a multi-layer composite sealed window glass structure, solving the problems of easy cracking and high-temperature gas leakage caused by thermal stress concentration in existing single-layer window glass. The double-layer silicone pad thermal compensation design can adapt to changes in high-temperature environment, ensuring long-term stable operation of the window glass and guaranteeing equipment safety.

[0060] 3. The detection system of this invention, through its bent base frame with grid holes and adjustable pitch mounting bracket design, achieves both efficient heat insulation and stray light shielding, while also allowing for flexible adjustment of the optical axis positions of the camera and light source, adapting to the detection needs of workpieces of different specifications and furnace structures. Simultaneously, the corrugated plate design on the furnace body facing the camera effectively reduces the surface temperature of the outer shell, minimizing the impact of heat radiation on the detection equipment and further improving the system's operational stability.

[0061] 4. This invention employs a multi-dimensional denoising preprocessing, adaptive background model updating, and a detection method that uses area, consecutive frame count, and motion direction as triple judgments. This effectively resists interference from complex environments such as furnace heat radiation, dust, and smoke, ensuring long-term detection accuracy. Simultaneously, the image tracing and automatic reset functions for fall events provide a basis for production anomaly analysis and ensure continuous stable operation, meeting the high reliability and high automation requirements of large-scale continuous production in industrial heat treatment furnaces. Attached Figure Description

[0062] Figure 1 This is a three-dimensional structural diagram of the workpiece drop detection system of the present invention.

[0063] Figure 2 This is a schematic diagram of the installation of the upper viewing window glass in the multi-layer composite sealing structure of the present invention.

[0064] Figure 3 This is a flowchart of the steps of the workpiece drop detection method of the present invention.

[0065] Figure 4 This is a schematic diagram of the background modeling process for step S1 of the workpiece drop detection method of the present invention.

[0066] Figure 5 This is a schematic diagram of the workpiece drop determination process in step S4 of the workpiece drop detection method of the present invention. Detailed Implementation

[0067] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0068] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0069] The technical solution of the present invention will be further described in detail below with reference to specific embodiments. These embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0070] like Figure 1 and Figure 2As shown, this embodiment provides a workpiece drop detection system based on the interior of a high-temperature furnace, including a furnace body 1. An upper cavity 2 is recessed in the upper part of the furnace body; a lower cavity 3 is recessed in the lower part of the furnace body, arranged vertically with the upper cavity 2, and the opening size and depth are completely consistent with the upper cavity. An upper viewing window 4 is sealed at the opening of the upper cavity 2, and a lower viewing window 5 is sealed at the opening of the lower cavity 3. A camera 6 is mounted on the outside of the upper cavity 2 via a first base frame 71, and detects workpieces falling into the furnace through the upper viewing window 4. A light source 8 is mounted on the outside of the lower cavity 3 via a second base frame 72, and provides illumination to the furnace through the lower viewing window 5. Specifically, the furnace body 1 can be formed by welding steel plates; the upper viewing window 4 and the lower viewing window 5 use the same high-temperature light-transmitting components; the camera 6 can be an industrial high-temperature resistant high-definition global shutter camera (e.g., model: MV-CA050-10GM), which can clearly capture images of workpieces inside the furnace through the viewing window; the light source 8 is a spotlight. The first base frame 71 and the second base frame 72 can both be made of 304 stainless steel and fixed to the furnace body to ensure a firm installation, while also possessing good high-temperature resistance and corrosion resistance, suitable for the high-temperature environment around the furnace body.

[0071] As described above, this invention creates an upper and lower cavity by recessing the upper and lower parts of the furnace body, respectively. The camera and light source are mounted on the outer sides of these cavities via a base frame, forming a dark-field illumination detection architecture where the camera detects from top to bottom and the light source provides supplementary illumination from bottom to top. This enables non-contact, real-time detection of workpieces falling inside the high-temperature furnace, solving problems inherent in existing technologies such as the susceptibility of contact sensors to high-temperature damage, severe interference from furnace thermal radiation under bright-field illumination leading to low contrast between the workpiece and background, and the extremely short lifespan of built-in detection equipment directly exposed to the high-temperature environment. This significantly improves the accuracy of workpiece identification and the overall lifespan of the system. Furthermore, the upper and lower cavity and window structure ensure precise matching between the detection field of view and the illumination area. This external installation design avoids direct exposure of the detection equipment to the high-temperature environment inside the furnace, ensuring stable operation of the camera and light source at safe temperatures and improving maintenance convenience and replacement efficiency. In addition, the modular design with upper and lower layers allows for independent debugging and maintenance of the camera or light source without shutting down and disassembling the furnace body, meeting the requirements of high reliability and low downtime for large-scale continuous production in industrial heat treatment furnaces.

[0072] like Figure 2As shown, in a preferred embodiment, the upper viewing window 4 adopts a multi-layer composite sealing structure, which is stacked sequentially from the inside of the furnace to the outside: inner pressure plate 41, first silicone pad 42, high-temperature tempered glass 43, second silicone pad 44, and outer pressure plate 45. The inner pressure plate 41 and the outer pressure plate 45 press the high-temperature tempered glass 43 tightly against the opening of the upper cavity 2 of the furnace body by bolts 46. The silicone pad serves as a seal and provides thermal compensation. The structure, size, and material of the lower viewing window 5 are completely identical to those of the upper viewing window 4, ensuring that the light source can stably penetrate and providing uniform dark-field illumination for the camera. Specifically, the inner pressure plate, outer pressure plate, first silicone pad, and second silicone pad are in a ring structure and will not obstruct the high-temperature tempered glass; the first and second silicone pads are 3mm thick high-temperature resistant silicone pads; the inner pressure plate and the outer pressure plate are 6mm thick 304 stainless steel plates. The high-temperature tempered glass 43 is preferably alkali-free aluminosilicate glass, with a softening temperature higher than 900℃ and a coefficient of thermal expansion of (42~46)×10⁻. 7 With a bending strength of 200~210MPa and tempered finish, it exhibits excellent thermal shock resistance and mechanical strength, making it suitable for high-temperature furnace environments below 800℃. The thickness of the high-temperature tempered glass 43 can be selected based on the furnace pressure, and the aperture size is determined according to the camera's field of view.

[0073] As described above, this invention, by designing the upper and lower viewing windows as a multi-layered composite sealing structure consisting of an inner pressure plate, a first silicone pad, high-temperature tempered glass, a second silicone pad, and an outer pressure plate stacked sequentially from the inside out, achieves high sealing performance and high thermal stability of the viewing windows. This solves the problems of easy cracking due to concentrated thermal stress, high-temperature gas leakage due to poor sealing, and frequent window replacement in existing single-layer viewing windows, thus significantly improving the service life of the viewing windows in high-temperature environments and the operational safety of the equipment. Simultaneously, the design of the double-layer silicone pads to compensate for the differences in thermal expansion and contraction of various components avoids the problem of compression cracking caused by the different expansion coefficients of glass and metal components in high-temperature environments, ensuring long-term stable light transmission and sealing performance of the viewing windows at high temperatures. Furthermore, the bolt-tightened detachable structure allows for quick cleaning and replacement of the viewing windows without the need for large-scale disassembly of the furnace body, reducing maintenance costs.

[0074] like Figure 1As shown, in a preferred embodiment, the first base frame 71 includes a first planar plate 711 extending into the upper cavity 2, and a second planar plate 712 formed by bending the first planar plate 711 downwards and extending outwards from the upper cavity 2. The first planar plate 711 is evenly provided with a plurality of first grid holes 713 for heat dissipation and reducing structural weight. The second planar plate 712 is used to mount the camera 6. The second base frame 72 includes a third planar plate 721 extending into the lower cavity 3, and a fourth planar plate 722 formed by bending the third planar plate 721 downwards and extending outwards from the lower cavity 3. The third planar plate 721 is evenly provided with a plurality of second grid holes 723, and the fourth planar plate 722 is used to mount the light source 8. In this embodiment, both the first base frame 71 and the second base frame 72 are made of 304 stainless steel, and the bending angle is 90°.

[0075] As described above, the base frame of this invention adopts a bent structure, which allows the camera and light source to be mounted on the outside of the furnace body, and increases rigidity through bending. Furthermore, the base frame is fixedly connected to the furnace body by extending into the furnace, resulting in a compact structure. In addition, the design of the grille holes reduces the weight of the base frame and increases air convection and heat dissipation area, thereby reducing the conduction of high temperatures from the furnace to the base frame, lowering the operating temperature of the camera and light source, preventing high-temperature damage to the equipment, while also allowing light to pass through.

[0076] like Figure 1 As shown, in a preferred embodiment, the camera 6 is mounted on the first base frame 71 via an adjusting bracket 61; the light source 8 is a spotlight, mounted on the second base frame 72 via an adjusting support 81; the adjusting bracket 61 is used to adjust the pitch angle of the camera 6, and the adjusting support 81 is used to adjust the pitch angle of the light source 7, so that their optical axes can be flexibly aligned with the falling workpiece; wherein, the camera 6 faces the upper cavity 2 and is tilted downwards via the adjusting bracket 61, and the light source 8 faces the lower cavity 3 and is tilted upwards via the adjusting support 81. In specific implementation, the adjusting bracket 61 is provided with an arc-shaped elongated hole, which, together with the locking bolt, allows for continuous adjustment of the pitch angle. The camera 6 is tilted downwards, and the light source 8 is tilted upwards, so that their optical axes intersect in the detection area inside the furnace, and the included angle can be adjusted within the range of 30°~60°.

[0077] As described above, this invention mounts the camera on a first base frame via an adjustable bracket and the light source on a second base frame via an adjustable support, both with adjustable tilt angles. This allows for flexible alignment of the camera and light source optical axes, solving the problems of fixed installation angles, inability to adapt to different workpiece sizes and detection areas, and decreased detection accuracy due to optical axis misalignment in existing technologies. This significantly improves the system's adaptability and detection accuracy under different working conditions. Furthermore, the downward tilt of the camera and the upward tilt of the light source ensure that their optical axes converge in the workpiece drop area within the furnace, allowing for clear capture and accurate identification of fallen workpieces. The adjustable bracket and support design enables rapid angle adjustment and fixation.

[0078] like Figure 1 As shown, the outer shell of the furnace body 1 is at least partially composed of corrugated steel plates. Preferably, the plate facing the camera of the furnace body 1 is a corrugated plate 101, while the plates of the remaining side walls are corrugated plates or flat steel plates. A corrugated plate is a sheet material with a corrugated cross-section formed by mechanical molding or bending. It utilizes the corrugation effect to significantly improve out-of-plane stiffness and stability, achieving a combination of lightweight and high strength. Specifically, the furnace body corrugated plate is a metal plate with a periodic corrugated shape, used to improve the structural strength of the furnace shell and absorb the thermal expansion deformation of the furnace body under high-temperature conditions.

[0079] As described above, this invention, by designing the plate facing the camera of the furnace body as a corrugated plate structure, achieves efficient heat dissipation and diffuse reflection of stray light from the furnace shell. This solves the problems in existing technologies where the high surface temperature of a flat steel plate shell leads to excessively high ambient temperatures, and severe specular reflection causes light spots on the viewing window to interfere with detection. This significantly improves the operational stability and image acquisition quality of the detection equipment. Simultaneously, the concave-convex structure of the corrugated plate increases the heat dissipation area of ​​the furnace shell compared to a flat steel plate of the same size, effectively reducing the total heat radiation power emitted by the furnace body. This prevents the camera from overheating and being damaged due to prolonged exposure to high temperatures, extending the equipment's lifespan. Furthermore, the diffuse reflection characteristics of the corrugated plate disperse external stray light such as workshop lights and sunlight in various directions, preventing concentrated specular reflection spots from entering the viewing window and interfering with camera imaging. This ensures a uniform background without obvious bright spots in the acquired image, improving the accuracy and stability of the detection. The design of placing the corrugated plate only on the side facing the camera balances heat dissipation, stray light suppression, and equipment manufacturing costs, requiring no modification to the overall furnace structure and adapting to the upgrade and modification needs of existing high-temperature furnaces.

[0080] The following table compares the heat dissipation effects of corrugated steel plates and flat steel plates (furnace temperature 1000℃, ambient temperature 25℃):

[0081] flat steel plate 1.8 125 1200 72 Corrugated board 2.4 (33% increase relative to the plane) 95 800 58

[0082] Example 2

[0083] like Figure 3 As shown in Embodiment 2 of this case, a workpiece drop detection method based on the interior of a high-temperature furnace is provided, which is applied to the workpiece drop detection system based on the interior of a high-temperature furnace described in Embodiment 1, and includes the following steps:

[0084] Step S1: Establish a baseline background model when no workpieces fall into the furnace;

[0085] Step S2: Under dark lighting conditions, acquire a continuous image sequence of the detection area inside the furnace;

[0086] Step S3: Compare each frame of the continuous image sequence with the reference background model in turn to extract candidate targets;

[0087] Step S4: Determine whether a workpiece has fallen based on the area, number of consecutive frames, and direction of motion of the candidate target.

[0088] Step S5: If it is determined that the workpiece has fallen, output the falling signal and / or the counting signal.

[0089] As described above, this invention establishes a baseline background model in step S1, acquires images under dark illumination in step S2, extracts candidate targets through background subtraction in step S3, determines workpiece drop based on multi-dimensional features in step S4, and constructs the entire workpiece drop detection method flow through the output signal in step S5. This enables automated real-time detection of workpiece drops in high-temperature furnaces, solving the problems of low efficiency, high missed detection rate, inability to respond in real time, and poor anti-interference capability of traditional image detection methods in existing technologies. This significantly improves the efficiency and real-time performance of workpiece drop detection. Simultaneously, the multi-dimensional judgment mechanism based on area, number of consecutive frames, and direction of motion enables the system to effectively distinguish between real workpiece drops and various interference signals. This detection principle, using dark illumination combined with background subtraction, avoids the influence of complex environmental factors such as furnace heat radiation and dust on the detection results, ensuring accurate identification of workpiece drop events and reducing false alarm and missed detection rates. Furthermore, the design of simultaneously outputting drop signals and piece counting signals allows for timely notification of operators to handle anomalies and automatic statistics of production data, meeting the management needs of automated production in industrial heat treatment furnaces.

[0090] like Figure 4 As shown, in a preferred embodiment, step S1 specifically includes:

[0091] Step S11: Power on camera 6 and light source 8, and allow 30 seconds of preheating after powering on to ensure stable operation of the equipment and stabilize the camera and light source.

[0092] Step S12: After confirming that no workpieces are falling into the furnace, continuously acquire the initial background image for a preset number of frames (e.g., 100 frames);

[0093] Step S13: Perform multi-dimensional denoising and dark field feature filtering on each frame of the initial background image in sequence;

[0094] Step S14: The preprocessed initial background image is fused at the pixel level using a multi-frame averaging method to obtain an initial reference background image B0(x,y), where (x,y) are the image pixel coordinates; at the same time, the gray standard deviation σ(x,y) of each pixel in the multi-frame preprocessed image is calculated and used as the background fluctuation tolerance threshold for that pixel, which is used to distinguish between background fluctuations and real targets in subsequent steps; the initial reference background image B0(x,y) and the gray standard deviation σ(x,y) data are stored to construct a reference background model.

[0095] Specifically, a multi-frame averaging method is used to perform pixel-level fusion on the 50 preprocessed initial background images. For any pixel (x, y) in the image, the gray value of the initial reference background image B0(x, y) is the average gray value of that pixel in the 50 preprocessed images. The calculation formula is as follows:

[0096] ;

[0097] Where I_i(x,y) is the gray value of the preprocessed image of the i-th frame at the (x,y) coordinates.

[0098] Simultaneously, the grayscale standard deviation σ(x,y) of each pixel (x,y) in the 50 preprocessed images is calculated using the following formula:

[0099]

[0100] The obtained grayscale standard deviation ranges from 2 to 8, which is used as the background fluctuation tolerance threshold for that pixel. For example, if σ(x,y)=5 for a certain pixel, then in subsequent detection, if the grayscale fluctuation of that pixel is ≤5, it is judged as background fluctuation; if it exceeds 5, it is judged as a possible target pixel.

[0101] Finally, the initial baseline background image B0(x,y) and the grayscale standard deviation σ(x,y) of all pixels are stored in the system memory to complete the baseline background model for comparison in the subsequent step S3.

[0102] As described above, this invention establishes a benchmark background model through a series of steps: powering on the device, acquiring multiple initial background images, performing multi-dimensional denoising preprocessing, constructing an initial benchmark background image by averaging and fusing multiple frames, and calculating the grayscale standard deviation as a fluctuation tolerance threshold. This enables the construction of a high-precision and highly stable benchmark background model, solving the problems in existing technologies where background models are easily affected by noise, cannot distinguish between background fluctuations and real targets, and have low initial background accuracy leading to large subsequent detection errors. This significantly improves the accuracy of subsequent target extraction and the stability of the detection system. Simultaneously, the pixel-level fusion design using multi-frame averaging effectively suppresses random noise in single-frame images. By introducing the grayscale standard deviation as the background fluctuation tolerance threshold for each pixel, the design avoids the problem of fixed thresholds being unable to adapt to local background fluctuations, ensuring accurate differentiation between subtle changes in the background and the real workpiece target, reducing false alarms. Furthermore, the design of uniformly storing background images and fluctuation threshold data to construct the model facilitates subsequent model calls and updates, improving system operating efficiency and maintainability.

[0103] In a preferred embodiment, step S13 includes:

[0104] Median filtering is used to remove salt-and-pepper noise generated by thermal radiation inside the furnace and transient interference from suspended dust particles;

[0105] Gaussian filtering is used to smooth local grayscale fluctuations caused by thermal stress deformation of the window glass;

[0106] Based on the principle of low grayscale background and high grayscale of fallen workpiece under dark lighting, an adaptive grayscale threshold T1 is set. Pixels with grayscale values ​​lower than the grayscale threshold T1 are retained as valid background pixels, while isolated pixels with grayscale values ​​higher than the grayscale threshold T1 and connected region area smaller than the preset minimum noise area S1 are identified as thermal noise points and removed.

[0107] In practice, a 3×3 filtering template is used. For each pixel in the initial background image of each frame, the grayscale values ​​of the pixel itself and its eight neighboring pixels are taken, sorted, and the median value is taken as the new grayscale value of the pixel. This removes salt-and-pepper noise (isolated pixels with abrupt grayscale value changes) caused by thermal radiation from the furnace and transient interference from suspended dust particles. For example, if the original grayscale value of a pixel is 150 (a salt-and-pepper noise point), and the grayscale values ​​of its eight surrounding pixels are all between 60 and 70, and the median value after sorting is 65, then the grayscale value of the pixel is replaced with 65, thus removing noise. A 5×5 Gaussian filtering template is used to smooth the image after median filtering, reducing local grayscale fluctuations caused by thermal stress deformation of the window glass (manifested as slight fluctuations in grayscale values ​​in local areas). For example, if a certain area of ​​the window glass experiences grayscale fluctuations due to thermal stress, with the grayscale value fluctuating between 55 and 75, after Gaussian filtering, the grayscale value of this area is smoothed to between 60 and 65, reducing the impact of fluctuations on the background model. Based on the characteristics of low background grayscale and high workpiece grayscale under dark lighting, the Otsu adaptive thresholding algorithm is adopted to automatically calculate the adaptive grayscale threshold T1=100. Pixels with grayscale values ​​below 100 are retained as valid background pixels (background grayscale values ​​range from 50-80, all below T1). A preset minimum noise area S1=20 pixels (corresponding to an actual area of ​​approximately 1.5mm²) is used to identify isolated pixels (thermal noise points, such as tiny dust particles or thermal radiation interference points) with grayscale values ​​above 100 and connected component areas less than 20 pixels as noise and remove them. For example, if an isolated pixel has a grayscale value of 110 and a connected component area of ​​15 pixels (<20 pixels), it is identified as a thermal noise point and removed. After removal, the grayscale value of this pixel is replaced with the average value of the surrounding background pixels. After preprocessing, the background grayscale of each frame is uniform, with no obvious noise or grayscale fluctuations, providing high-quality image data for subsequent baseline background model construction.

[0108] As described above, this invention achieves high-quality preprocessing of furnace background images by sequentially employing median filtering to remove salt-and-pepper noise and dust interference, Gaussian filtering to smooth grayscale fluctuations caused by thermal stress, and adaptive grayscale thresholding combined with connected component area filtering to eliminate thermal noise points. This multi-dimensional denoising and dark-field feature filtering process solves the problems in existing technologies where a single denoising method cannot simultaneously remove multiple types of interference and where uneven background grayscale makes subsequent target extraction difficult, thus significantly improving the quality of the preprocessed image and the accuracy of the background model. Furthermore, the combined use of median and Gaussian filtering allows for targeted removal of different types of noise. By setting an adaptive grayscale threshold based on dark-field illumination characteristics and combining it with connected component area filtering, the problem of small thermal noise points being misidentified as targets is avoided, ensuring that the preprocessed background image has uniform grayscale and no obvious interference, laying a solid foundation for the subsequent construction of a baseline background model. In addition, this preprocessing process has low computational load and high processing speed, meeting the efficiency requirements of real-time detection and adapting to the high-speed demands of industrial production.

[0109] In a preferred embodiment, step S1 further includes:

[0110] Step S15: During the detection process, if no workpiece falls within a preset time period (e.g., 60 seconds), the background model update mechanism is activated, and steps S12 and S13 are executed; then, the weighted average method is used to slowly update the reference background model, and the update formula is:

[0111] ;

[0112] Among them, B k (x,y) represents the updated k-th generation baseline background image, B k₋1 (x,y) represents the (k-1)th generation baseline background image before the update, B n (x,y) represents the average image of the n preprocessed frames collected in this acquisition, and α is the preset background update weight.

[0113] In practice, n=20 real-time images without workpieces are acquired (frame rate 20fps, acquisition time 1 second). These 20 images undergo the same preprocessing as in step S13 (3×3 median filtering, 5×5 Gaussian filtering, adaptive thresholding). The average image B of these 20 preprocessed images is then calculated. n (x,y) (Calculation method is the same as in step S14). The preset background update weight α = 0.9 (range 0.8-0.95; the closer α is to 1, the slower the update, ensuring the stability of the background model), is calculated according to update formula B. k (x,y)=0.9xB k₋1(x,y)+0.1xBn(x,y) is used to slowly update the baseline background model. For example, the grayscale value B of a certain pixel before the update. k₋1 If (x,y)=65, and the average grayscale value of this pixel in the collected image is Bn(x,y)=68, then the updated grayscale value B k (x,y)=0.9×65+0.1×68=65.3, achieving slow updates and avoiding sudden changes in the background model.

[0114] Step S16: If a candidate target is found in the pre-detection before background update, or if the overall gray value change of the window glass area exceeds the preset threshold T2 (indicating that the window is severely dusty or blocked by foreign objects), the current background update operation is immediately abandoned, and a window cleaning prompt signal is triggered. The background acquisition and update process is then re-executed after the anomaly is resolved.

[0115] In practice, before initiating the background update mechanism, a pre-detection is performed, acquiring three real-time images to extract candidate targets. Simultaneously, the overall grayscale value of the viewing window area is detected, and the overall average grayscale value is calculated. A preset threshold T2=30 (i.e., if the overall grayscale value changes by more than 30, it is determined that the viewing window is obstructed by dust or foreign objects) is set. If a candidate target (area 1500mm², appearing consecutively in three frames) is found during the pre-detection, the current background update is immediately abandoned, steps S12 and S13 and the update operation are not executed, and the original baseline background model is maintained until the candidate target disappears, then the system waits for the next 15-minute cycle to initiate the update. If no candidate target is found during the pre-detection, but the overall average grayscale value of the viewing window area changes from 65 to 100 (change 35 > T2=30), it is determined that the viewing window is severely dusty, the current update is immediately abandoned, and a viewing window cleaning prompt signal (audio-visual prompt) is triggered. After the operator cleans the viewing window, the prompt signal is manually reset, and the system re-executes the background acquisition and update process.

[0116] As described above, this invention, by initiating a background model update mechanism when no workpiece falls, employing a weighted average method for slow updates, and performing pre-detection and window grayscale changes before updates, achieves adaptive updates and anomaly protection for the baseline background model. This solves the problems in existing technologies where the background model cannot adapt to slow changes in the furnace environment, leading to decreased detection accuracy; the update process is easily contaminated by workpieces or foreign objects; and severe dust accumulation on the window cannot be detected in time. This significantly improves the long-term stability and detection accuracy of the system. Simultaneously, the slow update mechanism using the weighted average method allows the background model to gradually adapt to minor changes in the furnace environment without abrupt changes. The pre-detection before updates and the window grayscale anomaly detection design prevent the background model from being contaminated by fallen workpieces or obstructions, ensuring that the background model always accurately reflects the true background state inside the furnace. Furthermore, the design of triggering a window cleaning prompt signal promptly reminds operators to perform maintenance, preventing detection failure due to severe dust accumulation on the window, reducing equipment maintenance difficulty and downtime risk.

[0117] As a preferred embodiment, step S2, acquiring a continuous image sequence of the detection area inside the furnace under dark illumination conditions, includes:

[0118] Step S21: Light source 8 tilts upwards to illuminate the area inside the furnace, forming dark field lighting, so that the background area presents low grayscale and the workpiece surface presents high grayscale reflection characteristics;

[0119] Step S22: Camera 6 tilts from top to bottom to collect a continuous sequence of images inside the furnace in real time, denoted as I_t(x,y), where t represents the frame number; where t=1,2,3...

[0120] In a preferred embodiment, step S3 involves sequentially comparing each frame of the image sequence with a reference background model to extract candidate targets, including:

[0121] Step S31: Perform pixel-by-pixel difference comparison between the current frame image I_t(x,y) and the reference background image B(x,y), and calculate the difference image D_t(x,y)=|I_t(x,y)-B(x,y)|; the larger the gray value in the difference image, the more obvious the difference between it and the background.

[0122] Step S32: Based on the grayscale standard deviation σ(x,y) of each pixel in the baseline background model, set the adaptive difference threshold T_diff(x,y)=k·σ(x,y), where k is a preset scaling factor. In specific implementation, the preset scaling factor k=3 (range 2-4). For example, if σ(x,y)=5 for a certain pixel, then the difference threshold T_diff(x,y)=15 for that pixel. When the grayscale value of that pixel in the difference image is >15, it is determined to be a candidate target pixel.

[0123] Step S33: Binarize the difference image by marking the pixels where D_t(x,y)>T_diff(x,y) as candidate target pixels to obtain the binarized image F_t(x,y); in specific implementation, the pixels where D_t(x,y)>T_diff(x,y) are marked as 255 (white, candidate target pixels), and the remaining pixels are marked as 0 (black, background pixels) to obtain the binarized image F_t(x,y).

[0124] Step S34: Perform morphological opening operation on the binarized image F_t(x,y) using a 3×3 structuring element, first eroding and then dilating to remove isolated noise points and small interference areas (such as spots with a diameter of less than 3 pixels), while retaining larger candidate target areas.

[0125] Step S35: Perform connected component labeling on the binarized image after morphological processing (eight-connected component labeling method can be used), and extract the area, centroid position (such as centroid coordinates (x_c, y_c)) and circumscribed rectangle features (top left corner coordinates, width, and height) of each candidate target region.

[0126] like Figure 5 As shown, in a preferred embodiment, step S4, determining whether a workpiece has fallen based on the area, number of consecutive frames, and direction of motion of the candidate target, includes:

[0127] Step S41: Calculate the area A_i of each candidate target region. If the preset lower limit of the workpiece area is ≤A_i ≤ the preset upper limit of the workpiece area, it is retained as a valid candidate target. For example, the preset lower limit of the workpiece area is 2000 pixels (corresponding to an actual area of ​​about 2000 mm²) and the upper limit is 3000 pixels (corresponding to an actual area of ​​about 3000 mm²). Calculate the area A_i of each candidate target region. If A_i is within the range of 2000-3000 pixels, it is retained as a valid candidate target; otherwise, it is judged as noise and discarded.

[0128] Step S42: Perform target association tracking on candidate targets in a continuous frame sequence and record the number of consecutive frames C of the same target. In specific implementation, the Kalman filter algorithm can be used to perform target association tracking on candidate targets in a continuous frame sequence. Based on the centroid position, area and other features of the target, the position of the same target in different frames is matched and the number of consecutive frames C of the same target is recorded.

[0129] Step S43: When C ≥ the preset consecutive frame count threshold, proceed to the next step of judgment; for example, if the preset consecutive frame count threshold is 5 frames, when C ≥ 5, proceed to the next step of judgment; otherwise, it is judged as transient interference and is removed.

[0130] Step S44: Calculate the centroid displacement vector of the candidate target in consecutive frames and determine the angle θ between its motion direction and the direction of gravity. For example, if the centroid coordinates of a target in 5 frames are (1000, 500), (1005, 550), (1010, 600), (1015, 650), and (1020, 700), then its average motion direction is to the lower right, and the angle θ with the direction of gravity is θ = arctan(20 / 200) = 5.7°.

[0131] Step S45: When θ ≤ the preset maximum allowable deflection angle, the candidate target is determined to be a workpiece falling event; for example, if the preset maximum allowable deflection angle is 30°, when θ ≤ 30°, the candidate target is determined to be a workpiece falling event; otherwise, it is determined to be a non-falling target (such as a horizontally moving workpiece or floating objects in the furnace) and is rejected.

[0132] In a preferred embodiment, step S5, if it is determined that a workpiece has fallen, outputs a falling signal and / or a counting signal, including:

[0133] Step S51: When a workpiece falling event is determined, immediately output a falling alarm signal; at the same time, trigger an audible and visual alarm (red LED flashes and buzzer sounds an alarm).

[0134] Step S52: Simultaneously update the workpiece drop count value (cumulative count +1) and output the workpiece counting signal;

[0135] Step S53: Save the image data at the moment of the fall, including the current frame image and several frames before and after, for post-event retrospective analysis; in specific implementation, the image data at the moment of the fall can be stored in JPG format on the local hard drive, and the file name includes the fall timestamp.

[0136] Step S54: If no new falling event is detected within the preset time window, the detection state is automatically reset, and the system waits for the next detection. For example, if the preset time window is 30 seconds, and no new falling event is detected within 30 seconds, the detection state is automatically reset, the candidate target tracking list is cleared, and the system waits for the next detection.

[0137] Taking a workpiece with an area of ​​A_i=850 pixels falling through the detection area as an example: the preset lower limit of area A_min=50 and the upper limit of area A_max=5000; if C=6 consecutive frames (threshold N=5) appear within the range, and the angle of the movement direction θ=8° (threshold θ_max=20°), the system determines that the workpiece has fallen, outputs an alarm signal, and saves the time of the fall and the images of the 5 frames before and after it.

[0138] The following table compares the detection performance of different combinations of judgment dimensions:

[0139] Judging by area only 85% 12% 8% Area + Number of consecutive frames for judgment 92% 5% 3% Area + Number of consecutive frames + Direction of motion 99.2% 0.3% 0.5%

[0140] As described above, firstly, this solution utilizes dark-field illumination combined with an up-and-down tilting acquisition method to create extremely high contrast between the workpiece and the background, significantly improving the accuracy of target extraction. Secondly, the adaptive differential threshold setting can dynamically adjust according to background fluctuations, avoiding missed or false detections caused by fixed thresholds. Furthermore, the combination of morphological opening operations and connected component labeling effectively removes noise interference and accurately extracts the features of candidate targets. Simultaneously, the multi-dimensional judgment mechanism (area, number of consecutive frames, direction of motion) can effectively distinguish between real workpiece drops and various interferences, resulting in an extremely low false alarm rate. Moreover, complete signal output and data storage functions can promptly notify operators and provide a basis for post-event analysis. In addition, the automatic reset mechanism ensures continuous and stable system operation without frequent manual intervention. Thus, this solution achieves high-precision, high-reliability real-time detection of workpiece drops within a high-temperature furnace, solving the problems of poor adaptability of fixed differential thresholds, high false alarm rates from single-feature judgments, and the inability to trace drop events in existing technologies, thereby significantly improving the overall performance and practicality of the detection system.

[0141] Taking the detection of a square metal workpiece falling as an example, the specific workflow of the detection method of the present invention is as follows:

[0142] Step S1: Establish a baseline background model. After starting the system, first confirm that no workpieces have fallen into the furnace, the viewing window glass is unobstructed, and the light source and camera are working properly. Collect 30 frames of initial background images without workpieces. After preprocessing such as denoising and filtering, the initial baseline background model is obtained by fusion and stored in the system memory for subsequent comparison.

[0143] Step S2: Activate the light source (spotlight, tilted upwards at 25°) to create dark field illumination, making the background inside the furnace appear low grayscale (grayscale value 50-80) and the workpiece surface appear high grayscale (grayscale value 180-220); the camera (tilted downwards at 20°, frame rate 20fps, resolution 5 million pixels) acquires a continuous sequence of images of the detection area inside the furnace in real time, with each frame image size being 2592×1944 pixels, and the acquisition is continuous without interruption.

[0144] Step S3: Compare each frame of real-time image with the baseline background model pixel by pixel, calculate the difference image, set an adaptive difference threshold, binarize and morphologically process the difference image, extract candidate targets, remove noise interference, and retain areas that may be workpieces.

[0145] Step S4: The preset workpiece area range is 2000-3000mm², the threshold for consecutive occurrence of frames is 5 frames, and the maximum allowable deviation angle between the direction of motion and the direction of gravity is 30°; the extracted candidate targets are analyzed. If the area of ​​the candidate target is 2450mm² (within the preset range), it appears consecutively for 6 frames, and the angle between the direction of motion and the direction of gravity is 25° (≤30°), then it is determined that the workpiece has fallen.

[0146] Step S5: After determining that a workpiece has fallen, the system immediately outputs a fall alarm signal (audio and visual alarm, alarm volume ≥80dB, light flashing red), and updates the piece count (cumulative count +1), and outputs a piece count signal; saves the current frame image and the images of the previous and next 5 frames (a total of 11 frames) at the moment of the fall for later traceability; if no new fall event is detected within 30 seconds, the system automatically resets and waits for the next detection.

[0147] In summary, this invention discloses a workpiece drop detection system and method based on the interior of a high-temperature furnace. By combining a layered external dark-field illumination detection architecture, multi-layered composite sealed viewing windows, a bent base with grid holes, and a multi-dimensional intelligent detection algorithm, it achieves non-contact, high-precision, and high-reliability real-time detection of workpiece drops within the high-temperature furnace. This invention completely solves the problems of short lifespan, difficult maintenance, poor anti-interference ability, and high false alarm rate in existing detection equipment. It can effectively avoid equipment failures and production line downtime caused by workpiece drops, significantly improving the automation level and production efficiency of industrial heat treatment production, and has extremely high application value.

[0148] As stated above, this case protects a workpiece drop detection system based on the interior of a high-temperature furnace. All technical solutions that are the same as or similar to this case should be considered to fall within the scope of protection of this case.

Claims

1. A workpiece drop detection system based on the interior of a high-temperature furnace, characterized in that, include: Furnace body (1); The upper cavity (2) is recessed and formed in the upper part of the furnace body (1); The lower cavity (3) is recessed and formed in the lower part of the furnace body (1); The upper viewing window glass (4) is installed over the opening of the upper cavity (2); The lower viewing window glass (5) is installed over the opening of the lower cavity (3); The camera (6) is installed on the outside of the upper cavity (2) through the first base frame (71) to detect the workpiece falling into the furnace through the upper viewing window (4); The light source (8) is installed on the outside of the lower cavity (3) via the second base frame (72) to provide light to the furnace through the lower viewing window glass (5).

2. The workpiece drop detection system based on the interior of a high-temperature furnace according to claim 1, characterized in that, The upper window glass (4) adopts a multi-layer composite sealing structure, including an inner pressure plate (41), a first silicone pad (42), a high-temperature tempered glass (43), a second silicone pad (44), and an outer pressure plate (45) stacked sequentially from the inside to the outside; wherein, the inner pressure plate (41) and the outer pressure plate (45) press the high-temperature tempered glass (43) into the opening of the upper cavity (2) of the furnace body by bolts (46); the lower window glass (5) has the same structure as the upper window glass (4).

3. The workpiece drop detection system based on the interior of a high-temperature furnace according to claim 1, characterized in that, The first base frame (71) includes a first planar plate (711) extending into the upper cavity (2) and a second planar plate (712) formed by bending the first planar plate (711) downward and extending outward from the upper cavity (2). The first planar plate (711) is evenly provided with a plurality of first grid holes (713), and the second planar plate (712) is used for mounting the camera (6). The second base frame (72) includes a third planar plate (721) extending into the lower cavity (3) and a fourth planar plate (722) formed by bending the third planar plate (721) downward and extending it outward from the lower cavity (3). The third planar plate (721) is evenly provided with a plurality of second grid holes (723), and the fourth planar plate (722) is used for mounting the light source (8).

4. The workpiece drop detection system based on the interior of a high-temperature furnace according to any one of claims 1-3, characterized in that, The camera (6) is mounted on the first base frame (71) via an adjustment bracket (61); the light source (8) is a spotlight, mounted on the second base frame (72) via an adjustment support (81); the adjustment bracket (61) is used to adjust the pitch angle of the camera (6), and the adjustment support (81) is used to adjust the pitch angle of the light source (7), so that the optical axes of the two can be flexibly aligned with the falling workpiece; wherein, the camera (6) faces the upper cavity (2) and is tilted downward via the adjustment bracket (61), and the light source (8) faces the lower cavity (3) and is tilted upward via the adjustment support (81).

5. The workpiece drop detection system based on the interior of a high-temperature furnace according to claim 1, characterized in that, The outer shell of the furnace body (1) includes at least one side wall facing the camera (6) which is a corrugated plate (101).

6. A method for detecting workpiece falling from inside a high-temperature furnace, characterized in that, The workpiece drop detection system based on the interior of a high-temperature furnace body as described in any one of claims 1 to 5 includes the following steps: Step S1: Establish a baseline background model when no workpieces fall into the furnace; Step S2: Under dark lighting conditions, acquire a continuous image sequence of the detection area inside the furnace; Step S3: Compare each frame of the continuous image sequence with the reference background model in turn to extract candidate targets; Step S4: Determine whether a workpiece has fallen based on the area, number of consecutive frames, and direction of motion of the candidate target. Step S5: If it is determined that the workpiece has fallen, output the falling signal and / or the counting signal.

7. The workpiece drop detection method based on the interior of a high-temperature furnace according to claim 6, step S1 includes: Step S11: Power on the camera (6) and the light source (8); Step S12: Under the condition that no workpieces fall, continuously acquire the initial background image inside the furnace for a preset number of frames; Step S13: Perform multi-dimensional denoising and dark field feature filtering on each frame of the initial background image in sequence; Step S14: The preprocessed initial background image is fused at the pixel level using the multi-frame averaging method to obtain the initial reference background image B0(x,y), where (x,y) are the image pixel coordinates; at the same time, the gray standard deviation σ(x,y) of each pixel in the multi-frame preprocessed image is calculated and used as the background fluctuation tolerance threshold of the pixel, which is used to distinguish the background fluctuation from the real target in subsequent steps. The initial baseline background image B0(x,y) and grayscale standard deviation σ(x,y) data are stored to construct the baseline background model.

8. The workpiece drop detection method based on the interior of a high-temperature furnace according to claim 7, wherein step S13 includes: Median filtering is used to remove salt-and-pepper noise generated by thermal radiation inside the furnace and transient interference from suspended dust particles; Gaussian filtering is used to smooth local grayscale fluctuations caused by thermal stress deformation of the window glass; Based on the principle of low grayscale background and high grayscale of fallen workpiece under dark lighting, an adaptive grayscale threshold T1 is set. Pixels with grayscale values ​​lower than the grayscale threshold T1 are retained as valid background pixels, while isolated pixels with grayscale values ​​higher than the grayscale threshold T1 and connected region area smaller than the preset minimum noise area S1 are identified as thermal noise points and removed.

9. The workpiece drop detection method based on the interior of a high-temperature furnace according to claim 7 or 8, step S1 further includes: Step S15: During the detection process, if no workpiece falls within the preset time limit, the background model update mechanism is activated, and steps S12 and S13 are executed; then, the weighted average method is used to slowly update the reference background model, and the update formula is: B k (x,y)=α×B k₋1 (x,y)+(1-α)×B n (x,y); Among them, B k (x,y) represents the updated k-th generation baseline background image, B k₋1 (x,y) represents the (k-1)th generation baseline background image before the update, B n (x,y) represents the average image of the n preprocessed frames collected in this acquisition, and α is the preset background update weight; Step S16: If a candidate target is found in the pre-detection before background update, or if the overall gray value change of the window glass area exceeds the preset threshold T2, the current background update operation is immediately abandoned, and a window cleaning prompt signal is triggered. The background acquisition and update process is then re-executed after the anomaly is resolved.

10. The workpiece drop detection method based on the interior of a high-temperature furnace according to claim 7 or 8, wherein step S2 includes: The light source (8) tilts upwards to illuminate the area inside the furnace, creating a dark field illumination. The camera (6) tilts from top to bottom to acquire a continuous image sequence I_t(x,y) in real time, where t represents the frame number; Step S3 includes: performing a pixel-by-pixel difference comparison between the current frame image I_t(x,y) and the reference background image B(x,y), and calculating the difference image D_t(x,y)=|I_t(x,y)-B(x,y)|; Based on the grayscale standard deviation σ(x,y) of each pixel in the baseline background model, an adaptive difference threshold T_diff(x,y) = k·σ(x,y) is set, where k is a preset scaling factor. The difference image is binarized, and pixels where D_t(x,y) > T_diff(x,y) are marked as candidate target pixels, resulting in a binarized image F_t(x,y). Morphological opening operations are performed on the binarized image F_t(x,y), first eroding and then dilating, to remove isolated noise points and small interference regions. Connected component labeling is performed on the morphologically processed binarized image, and the area, centroid position, and circumscribed rectangle features of each candidate target region are extracted. Step S4 includes: calculating the area A_i of each candidate target region; retaining a candidate target as valid when the preset lower limit of the workpiece area ≤ A_i ≤ the preset upper limit of the workpiece area; performing target association tracking on candidate targets in a continuous frame sequence and recording the number of consecutive frames C of the same target; proceeding to the next step when C ≥ the preset threshold for the number of consecutive frames; calculating the centroid displacement vector of the candidate target in the continuous frames and determining the angle θ between its motion direction and the direction of gravity. When θ ≤ the preset maximum allowable deflection angle, the candidate target is determined to be a workpiece falling event; Step S5 includes: outputting alarm signals and piece counting signals, saving the drop image, and automatically resetting if no new events occur within a preset time window.

Images