Intelligent feeding and sorting device with high-precision visual recognition

Abstract

This invention discloses a high-precision visual recognition intelligent feeding and sorting device, including a feeding conveyor line, a visual recognition station, a sorting execution mechanism, a control cabinet, and a support and adjustment structure. The feeding conveyor line has a guiding and sorting structure at its inlet end. The visual recognition station includes a gantry bracket and an industrial camera and a ring light source mounted on it. The sorting execution mechanism includes multiple sets of push rod cylinders and multiple material dispensing chutes. The control cabinet integrates a visual processing module and a motion control module. This method acquires material images through an industrial camera, and the visual processing module performs image preprocessing, segmentation, multi-dimensional feature extraction, and classification recognition to generate sorting control signals that drive the push rod cylinders, achieving automatic material sorting. Stable illumination is provided by the ring light source, and the integration of visual processing and motion control achieves an automated closed loop of "acquisition—recognition—decision—execution," offering advantages such as high recognition accuracy, high sorting efficiency, and a high degree of automation.

Description

Technical Field

[0001] This invention belongs to the field of intelligent sorting equipment technology, specifically relating to an intelligent feeding and sorting device with high-precision visual recognition. Background Technology

[0002] In modern industrial production, material sorting is a crucial process in product assembly, quality inspection, and packaging. Traditional material sorting mainly relies on manual visual identification and classification, which suffers from high labor intensity, low sorting efficiency, and unstable identification accuracy. Especially in the sorting of small-sized materials such as electronic components and precision hardware, the materials are diverse and have high similarity in appearance, making it difficult for the human eye to quickly and accurately distinguish them, easily leading to false positives and false negatives.

[0003] With the rapid development of machine vision technology, automated sorting equipment based on vision recognition is gradually being applied in industrial production. However, existing vision sorting equipment still has the following shortcomings: The lighting systems of existing vision sorting equipment mostly use fixed light sources or natural light, which leads to unstable lighting conditions and fluctuations in the image quality captured by industrial cameras, affecting the accuracy of the recognition algorithm. The recognition effect is particularly noticeable when there is reflection or shadow on the material surface.

[0004] Although some equipment is equipped with a visual recognition system, the sorting process after recognition still relies on manual operation or semi-automatic methods. There is a lack of automated linkage between visual recognition and sorting execution, which cannot form a complete closed loop from "image acquisition → recognition and classification → automatic sorting", resulting in limited improvement in sorting efficiency.

[0005] Existing vision sorting equipment is mostly customized, with scattered functional modules, large equipment size, and requires professional personnel for installation and commissioning, making it difficult to adapt to the rapid changeover needs of different production lines and different materials.

[0006] The existing equipment lacks an effective guiding and sorting structure in the feeding process, and materials are prone to stacking, shifting or tilting on the conveyor belt, making it difficult for the visual recognition station to accurately capture material images, thus affecting the recognition rate and sorting accuracy.

[0007] Existing sorting equipment has a limited number of sorting channels and can usually only achieve two-category sorting of good and bad products, which is difficult to meet the needs of multi-variety fine sorting of multiple materials. Summary of the Invention

[0008] In view of the shortcomings of the existing technology, the present invention aims to solve the technical problems of existing material sorting methods, such as the recognition accuracy being greatly affected by light, the lack of linkage between visual recognition and sorting execution, the lack of organization in the feeding process, and the limited sorting channels, and provides a high-precision visual recognition intelligent feeding and sorting device.

[0009] The technical solution adopted by this invention to solve its technical problem is: A high-precision visual recognition intelligent feeding and sorting device includes: A feeding conveyor line, including a frame and a horizontal conveyor belt disposed on the frame; A visual recognition station is set in the middle section of the feeding conveyor line, including a gantry support and an industrial camera and a ring light source installed on the top of the gantry support. The industrial camera is used to collect image information of the material, and the ring light source is used to provide illumination for image acquisition. The sorting execution mechanism is located at the end of the feeding conveyor line and includes multiple sets of push rod cylinders and multiple material dispensing slides. Each push rod cylinder is correspondingly set with each material dispensing slide. A control cabinet is located on the side of the frame and integrates a vision processing module and a motion control module. The vision processing module is electrically connected to the industrial camera, and the motion control module is electrically connected to the push rod cylinder. A support and adjustment structure is provided at the bottom of the frame, including multiple adjustable feet for adjusting the levelness of the equipment.

[0010] Preferably, the guiding and sorting structure includes guide plates symmetrically arranged on both sides of the feed end of the frame, and the distance between the two guide plates gradually narrows along the conveying direction.

[0011] Preferably, the gantry support spans over the horizontal conveyor belt, the industrial camera is positioned at the top center of the gantry support with its lens facing downwards, and the ring light source is arranged around the outer periphery of the industrial camera lens.

[0012] Preferably, the visual recognition station also includes a photoelectric sensor, which is located upstream of the industrial camera and is used to detect whether the material has arrived at the visual recognition station and trigger the industrial camera to acquire images.

[0013] Preferably, the vision processing module has a built-in image processing algorithm for extracting features and classifying the acquired images, and outputting sorting control signals to the motion control module.

[0014] Preferably, multiple material discharging chutes are arranged side by side along the width of the frame, with the inlet end of each chute corresponding to the end of the horizontal conveyor belt, and the outlet end of each chute extending to different material collection areas.

[0015] Preferably, the sorting execution mechanism further includes a sorting table, which is disposed between the end of the horizontal conveyor belt and the inlet end of the material dispensing chute, for receiving materials falling from the end of the horizontal conveyor belt; the push rod cylinders are respectively disposed at each sorting station of the sorting table, for pushing materials into the corresponding material dispensing chute.

[0016] Preferably, the control cabinet is equipped with a human-machine interface for displaying material identification results, sorting statistics and equipment operating status, and for receiving user operation commands.

[0017] Preferably, there are four adjustable feet, which are respectively located at the four corners of the bottom of the frame. Each adjustable foot includes a screw, an adjusting nut threadedly connected to the screw, and a base plate fixed to the bottom of the screw.

[0018] Preferably, the sorting method of the device includes the following steps: Step S1: The material is guided into the horizontal conveyor belt by the guiding and sorting structure, and passes through the vision recognition station in sequence along the conveying direction; Step S2: When the material arrives at the vision recognition station, the ring light source is lit to provide illumination, and the industrial camera captures the image information of the material and transmits it to the vision processing module of the control cabinet; Step S3: The vision processing module extracts and classifies features from the acquired images to determine the category or quality of the materials and generates corresponding sorting control signals. Step S4: The motion control module controls the corresponding pusher cylinder to move according to the sorting control signal, pushing the material into the corresponding material discharge chute to complete the sorting operation.

[0019] The beneficial effects of this invention are as follows: A ring light source is installed around the industrial camera lens at the vision recognition station. This ring light emits uniform, shadowless illumination, providing stable and constant lighting conditions for material image acquisition. This effectively eliminates glare and shadow interference from the material surface, significantly improving image quality and ensuring the stable operation of the vision recognition algorithm under high-precision requirements, achieving an accuracy rate of over 99%. The vision processing module and motion control module are integrated in the same control cabinet. After the industrial camera acquires images, the vision processing module performs feature extraction and classification in real time. The recognition results directly drive the corresponding push rod cylinder to automatically push the material into the corresponding material distribution chute. The entire process from image acquisition to sorting execution requires no manual intervention, forming an automated closed loop of "acquisition—recognition—decision—execution," greatly improving sorting efficiency.

[0020] A guiding and sorting structure is installed at the feeding end, with the spacing between the guide plates on both sides gradually narrowing along the conveying direction. This guides the materials to automatically center and align as they enter the horizontal conveyor belt, preventing material stacking, offset, or skew. This ensures that each piece of material enters the vision recognition station in a stable posture, improving recognition success rate and sorting accuracy. At the end, multiple sets of pusher cylinders and multiple material distribution and discharge chutes are arranged, with each pusher cylinder corresponding to a specific material distribution and discharge chute. Based on the material's identification and classification results, it can push the material into the corresponding chute, achieving fine classification and sorting of various materials and meeting the flexible sorting needs of multi-variety, small-batch production scenarios. A photoelectric sensor is installed upstream of the industrial camera. The camera is only triggered to acquire images when materials arrive at the vision recognition station, avoiding invalid data processing and resource waste caused by continuous photography, thus improving system operating efficiency and response speed. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the structure of a high-precision visual recognition intelligent feeding and sorting device provided in an embodiment of the present invention. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments are not intended to limit the present invention.

[0023] Example like Figure 1 As shown, a high-precision visual recognition intelligent feeding and sorting device includes a feeding conveyor line 10, a visual recognition station 20, a sorting execution mechanism 30, a control cabinet 40, and a support and adjustment structure 50.

[0024] The feeding conveyor line 10 includes a frame 11 and a horizontal conveyor belt 12 mounted on the frame 11. The frame 11 is a long, rectangular frame structure, welded from aluminum alloy profiles or square steel tubes, extending horizontally and possessing sufficient structural strength and rigidity. The length of the frame 11 is 1500-3000 mm, and the width is 300-600 mm, with the specific dimensions determined according to the material specifications and production cycle.

[0025] The horizontal conveyor belt 12 is a flat conveyor belt made of food-grade PVC or PU material, with an anti-slip texture on the surface to prevent materials from slipping during conveying.

[0026] The visual recognition station 20 is located in the middle section of the feeding conveyor line 10 and is used for image acquisition and visual recognition of the materials being conveyed. The visual recognition station 20 includes a gantry support 21, an industrial camera 22, a ring light source 23, and a photoelectric sensor.

[0027] The gantry support 21 spans across the horizontal conveyor belt 12 and is fixedly installed on both sides of the frame 11. The gantry support 21 has an inverted U-shaped structure, made of aluminum alloy profiles or square steel tubing, and possesses sufficient structural rigidity to support the weight of the industrial camera 22 and the ring light source 23. The top crossbeam 211 of the gantry support 21 extends along the width of the horizontal conveyor belt 12, and the height of the crossbeam 211 is 200-400mm above the surface of the horizontal conveyor belt 12.

[0028] Industrial camera 22 is mounted at the center of the top beam 211 of the gantry support 21, with the lens facing downwards and directly opposite the conveying surface of the horizontal conveyor belt 12. Industrial camera 22 is a high-definition industrial camera with a resolution of no less than 20 megapixels and a frame rate of no less than 30fps. It employs a global shutter sensor, enabling it to clearly capture images of materials moving at high speeds. The lens of industrial camera 22 is either a fixed-focus or zoom lens, with the focal length selected according to the field of view to ensure that the material image occupies 30%-70% of the screen area.

[0029] A ring light source 23 is arranged around the lens of the industrial camera 22 and fixed to the bottom of the crossbeam 211 of the gantry bracket 21 by a bracket. The ring light source 23 is a high-brightness LED ring light source, consisting of multiple LED beads arranged in a ring, emitting uniform, shadowless white or red light. The color temperature of the ring light source 23 is 5000-6500K, and its brightness is adjustable, providing stable and constant lighting conditions for material image acquisition, effectively eliminating reflections and shadows on the material surface, and improving image quality. The power cord of the ring light source 23 is electrically connected to the control cabinet 40, and its switching and brightness adjustment are controlled by the control system.

[0030] The photoelectric sensor is positioned upstream of the industrial camera 22 (i.e., in front of it in the material conveying direction), and is installed on the front side of the gantry bracket 21 or on both sides of the frame 11. The photoelectric sensor is either a through-beam or reflective type, used to detect whether material has reached the visual recognition station. When material passes through the detection area of ​​the photoelectric sensor, the sensor sends a trigger signal to the control cabinet 40, controlling the industrial camera 22 to acquire an image.

[0031] The distance between the photoelectric sensor and the industrial camera 22 is 50-150mm to ensure that the camera has completed focusing and exposure preparation when the material arrives directly below the industrial camera 22.

[0032] The sorting execution mechanism 30 is located at the end of the feeding conveyor line 10 and is used to automatically sort materials based on visual recognition results. The sorting execution mechanism 30 includes a sorting table 31, multiple sets of push rod cylinders 32, and multiple material dispensing chutes 33.

[0033] The sorting table 31 is located at the end of the horizontal conveyor belt 12 and is used to receive materials falling from the end of the horizontal conveyor belt 12. The sorting table 31 is a horizontally arranged rectangular platform made of metal sheet with a smooth upper surface that is flush with or slightly lower than the conveying surface of the horizontal conveyor belt 12. The sorting table 31 has multiple sorting stations 311, which are arranged along the width of the horizontal conveyor belt 12. Each sorting station 311 corresponds to a push rod cylinder 32 and a material dispensing chute 33.

[0034] Multiple sets of push rod cylinders 32 are respectively installed at each sorting station 311 of the sorting table 31, with each push rod cylinder 32 corresponding to a material dispensing chute 33. Each push rod cylinder 32 is a double-acting cylinder, including a cylinder body and a piston rod, with a push plate 321 at the end of the piston rod. The push plate 321 is a rectangular or circular flat plate made of plastic or metal, and its width matches the size of the material. When the piston rod of the push rod cylinder 32 extends, the push plate 321 pushes the material located at the sorting station 311 into the corresponding material dispensing chute 33.

[0035] Multiple material dispensing chutes 33 are arranged side-by-side along the width of the frame 11. The inlet end of each material dispensing chute 33 is connected to a corresponding sorting station 311 on the sorting table 31, and the outlet end of each material dispensing chute 33 extends to a different material collection area. The material dispensing chutes 33 are inclined chutes with an inclination angle of 15°-30°, allowing materials to automatically slide down into the collection container under gravity. The material dispensing chutes 33 are separated from each other by partitions 331 to prevent material mixing during the sliding process.

[0036] The number of material sorting and discharge chutes 33 is determined according to the sorting category. In this embodiment, four chutes are set to achieve sorting of four types of materials. As an alternative, the material sorting and discharge chutes 33 can also be set to two, three, or six chutes to adapt to different sorting needs.

[0037] The control cabinet 40 is located on the side of the frame 11 and integrates a vision processing module and a motion control module. The control cabinet 40 is a closed box structure made of metal sheet, which has good electromagnetic shielding and heat dissipation performance.

[0038] The vision processing module is a high-performance industrial computer or embedded AI computing platform, electrically connected to the industrial camera 22. It has built-in image processing algorithms for feature extraction and classification of acquired material images. The image processing flow of the vision processing module includes: image preprocessing (denoising and enhancement), image segmentation (target extraction), feature extraction (shape, color, texture, etc.), classification and recognition (comparison with preset templates or deep learning models), and result output.

[0039] The motion control module, a programmable logic controller or motion control card, is electrically connected to the solenoid valve of the push rod cylinder 32. It receives sorting control signals output from the vision processing module and controls the corresponding push rod cylinder 32 to move. The motion control module is also electrically connected to the conveyor motor 15, the ring light source 23, and photoelectric sensors to coordinate and control the timing of the actions of each actuator.

[0040] The control cabinet 40 is equipped with a human-machine interface, which is a touch screen installed on the front panel of the control cabinet 40. The human-machine interface is used to display material identification results, sorting statistics (such as the quantity of each type of material, sorting success rate, etc.) and equipment operating status, and to receive user operation commands, such as parameter setting, mode switching, manual control, etc.

[0041] The control cabinet 40 is also equipped with a power supply module, which is a switching power supply that converts external AC power into DC power required by each module, providing power to the industrial camera 22, ring light source 23, photoelectric sensor, vision processing module, motion control module and human-machine interface.

[0042] The support and adjustment structure 50 is located at the bottom of the frame 11.

[0043] The operator places the material to be sorted at the feed end of the horizontal conveyor belt 12. The material moves forward under the drive of the horizontal conveyor belt 12.

[0044] The sorted material continues to move forward with the horizontal conveyor belt 12. When the material reaches the visual recognition station 20, the photoelectric sensor detects the arrival of the material and sends a trigger signal to the control cabinet 40.

[0045] Upon receiving a trigger signal, the control cabinet 40 illuminates the ring light source 23 to provide stable illumination for image acquisition, and simultaneously triggers the industrial camera 22 to acquire image information of the material. The industrial camera 22 then transmits the acquired image to the vision processing module of the control cabinet 40.

[0046] The vision processing module preprocesses, segments, extracts features, and classifies the acquired images to determine the category or quality of the materials (such as qualified / unqualified, A-class / B-class / C-class, etc.) and generates corresponding sorting control signals.

[0047] The material continues to move forward and falls from the end of the horizontal conveyor belt 12 into the corresponding sorting station 311 on the sorting table 31. The motion control module activates the corresponding push rod cylinder 32 according to the sorting control signal output by the vision processing module.

[0048] The piston rod of the pusher cylinder 32 extends, and the pusher plate 321 pushes the material on the sorting station 311 toward the inlet end of the corresponding material discharge chute 33. The material slides down the material discharge chute 33 into the corresponding collection container, completing the sorting operation.

[0049] The human-machine interface displays the material identification results, sorting statistics, and equipment operating status in real time. Operators can monitor equipment operation through the human-machine interface and adjust equipment parameters as needed.

[0050] The image processing algorithm built into the visual processing module includes an image preprocessing unit, an image segmentation unit, a feature extraction unit, and a classification and recognition unit.

[0051] The raw images I(x,y) captured by industrial cameras are affected by factors such as environmental noise, material surface reflection, and conveyor line vibration, and need to be preprocessed to improve image quality.

[0052] Industrial cameras capture RGB color images, which are first converted to grayscale images to reduce computational complexity. A weighted grayscale conversion formula is used: G(x,y)=0.299×R(x,y)+0.587×G(x,y)+0.114×B(x,y); Where R(x,y), G(x,y), and B(x,y) are the pixel values ​​of the red, green, and blue channels, respectively.

[0053] Median filtering is used to remove salt-and-pepper noise and random noise from the image while effectively preserving material edge information. G_filtered(x,y)=median{G(x+i,y+j)}, (i,j)∈W; Where W is a 3×3 or 5×5 filter window, the window size is adjusted to adapt to the noise level of different materials. For motion ambiguity caused by conveyor line vibration, Wiener filtering is used for restoration.

[0054] To address potential overexposed or underexposed areas under ring light illumination, adaptive histogram equalization is used to enhance contrast. G_enhanced(x,y)=CLAHE(G_filtered(x,y)); The CLAHE algorithm effectively enhances the contrast between the material and the background by performing histogram equalization in local areas of the image and limiting the contrast amplification factor, while avoiding excessive noise amplification.

[0055] The preprocessed image needs to be segmented to separate the material area from the background.

[0056] The optimal segmentation threshold T is obtained using the Otsu global thresholding method. : T =argmax{σ_B^2(T)}; where σ_B^2(T) is the between-class variance, and T is the grayscale threshold. For uneven illumination conditions, adaptive local threshold segmentation is adopted, the image is divided into multiple sub-regions, and the optimal segmentation threshold of each region is calculated separately.

[0057] The segmented binary image B(x,y) may have holes or edge burrs, which are corrected by morphological opening and closing operations: B_open=(B⊖K)⊕K; B_close=(B⊕K)⊖K; where K is a structural element (usually a 3×3 or 5×5 circular or square structural element), ⊖ represents an erosion operation, and ⊕ represents a dilation operation. The opening operation is used to remove small noise points and disconnect narrow connections, and the closing operation is used to fill small holes and smooth edges.

[0058] Connected component labeling is performed on the processed binary image to extract the region of interest (ROI) of each material. The Two-Pass algorithm is used to label the connected components, and the bounding rectangle coordinates (x1,y1,x2,y2), area A, perimeter P and centroid coordinates (x_c,y_c) of each connected component are calculated.

[0059] Screening effective material regions: the area A is within a preset range (A_min<A<A_max), and the circularity C=4πA / P²; is within a reasonable range (0.6<C<1.2).

[0060] Multi-dimensional features are extracted from the segmented material ROI for subsequent classification and recognition.

[0061] Extract shape descriptors of materials: (1) Aspect ratio: R_aspect=W / H, where W and H are the width and height of the bounding rectangle, respectively.

[0062] (2) Rectangularity: R_rect=A / (W×H), which reflects the degree of proximity between the material and a rectangle.

[0063] (3) Circularity: C=4πA / P², which reflects the degree of proximity between the material and a circle.

[0064] (4) Hu moments: 7 Hu invariant moments (M1~M7) are calculated, which have rotation, translation and scale invariance.

[0065] Extract color features in RGB and HSV color spaces: (1) RGB mean: μ_R, μ_G, μ_B; (2) RGB standard deviation: σ_R, σ_G, σ_B; (3) Mean HSV values: μ_H, ​​μ_S, μ_V; (4) Color histogram: Each channel is quantized into 32 bins to form a 96-dimensional color histogram feature.

[0066] Texture features are extracted using the gray-level co-occurrence matrix: (1) Contrast: Contrast = ∑_{i,j}|ij|²P(i,j); (2) Correlation: Correlation=∑_{i,j}((i-μ_i)(j-μ_j)P(i,j)) / (σ_iσ_j); (3) Energy: Energy = ∑_{i,j}P(i,j)²; (4) Homogeneity: Homogeneity = ∑_{i,j}P(i,j) / (1+|ij|); Where P(i,j) is the gray-level co-occurrence matrix, i and j are gray-level indexes, μ_i and σ_i are the mean and standard deviation in the row direction, respectively, and μ_j and σ_j are the mean and standard deviation in the column direction, respectively. The average values ​​of the features in the four directions of 0°, 45°, 90°, and 135° are selected as the final texture features.

[0067] For complex materials, a lightweight convolutional neural network is used to extract deep features. The network structure includes: (1) Input layer: 128×128×3 ROI image; (2) Convolutional layer 1: 32 3×3 convolutional kernels, stride 1, ReLU activation, 2×2 max pooling; (3) Convolutional layer 2: 64 3×3 convolutional kernels, stride 1, ReLU activation, 2×2 max pooling; (4) Convolutional layer 3: 128 3×3 convolutional kernels, stride 1, ReLU activation, 2×2 max pooling; (5) Global average pooling layer; (6) Fully connected layer: 128-dimensional feature vector; The extracted shape features (12-dimensional), color features (102-dimensional), texture features (4-dimensional), and CNN depth features (128-dimensional) are fused to form a 246-dimensional comprehensive feature vector.

[0068] First, Z-score standardization is performed on each type of feature: z_i = (x_i - μ_i) / σ_i; After standardization, the feature vectors are concatenated into a comprehensive feature vector F=[F_shape; F_color; F_texture;F_cnn].

[0069] Multi-class support vector machine is used for classification. For K types of materials, a one-to-one strategy is used to construct K(K-1) / 2 binary classifiers.

[0070] The decision function of the SVM classifier is: f(x)=sign(∑_{i=1}^{n}α_i y_i K(x_i,x)+b); Where K(x_i,x) is the radial basis kernel function: K(x_i,x)=exp(-γ||x_i-x||²); γ = 1 / (2σ²), and the hyperparameter γ and the penalty parameter C are optimized through cross-validation.

[0071] The classifier outputs a decision value d_k for each class, which is then converted into a probability using Platt scaling. P(y=k|x)=1 / (1+exp(A_k·d_k+B_k)); Where A_k and B_k are scaling parameters obtained through maximum likelihood estimation. When the maximum probability P_max ≥ 0.85, the classification result is output; when P_max < 0.85, it is determined that the identification is uncertain, and the material is sent to the re-inspection channel.

[0072] The vision processing module outputs sorting control signals to the motion control module. The signal format includes: (1) Material ID: Uniquely identifies the current material; (2) Classification label k: 0~K-1, corresponding to different material distribution and discharge chutes; (3) Confidence score P_max: probability value between 0 and 1; (4) Sorting execution time T_action: The timing of triggering the push rod cylinder is estimated based on the time it takes for the material to arrive at the sorting table; After receiving the sorting control signal, the motion control module activates the corresponding push rod cylinder according to the classification label k to complete the automatic sorting operation.

[0073] Using the algorithm described above, this embodiment achieves a recognition accuracy of 99.2% under standard testing conditions, with a single recognition time of less than 50ms, meeting the real-time requirements of high-speed sorting production lines. For new material categories, incremental learning is supported—operators annotate new category samples through the human-machine interface, and the system automatically updates the classifier model parameters without requiring downtime for retraining, effectively improving the equipment's adaptability to new materials.

[0074] The above embodiments of the present invention are not intended to limit the scope of protection of the present invention. The implementation of the present invention is not limited thereto. All other modifications, substitutions or alterations made to the above structure of the present invention based on the above content of the present invention, in accordance with ordinary technical knowledge and common practice in the field, without departing from the basic technical idea of ​​the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A high-precision visual recognition intelligent feeding and sorting device, characterized in that, include: A feeding conveyor line, including a frame and a horizontal conveyor belt disposed on the frame; A visual recognition station is set in the middle section of the feeding conveyor line, including a gantry support and an industrial camera and a ring light source installed on the top of the gantry support. The industrial camera is used to collect image information of the material, and the ring light source is used to provide illumination for image acquisition. The sorting execution mechanism is located at the end of the feeding conveyor line and includes multiple sets of push rod cylinders and multiple material dispensing slides. Each push rod cylinder is correspondingly set with each material dispensing slide. A control cabinet is located on the side of the frame and integrates a vision processing module and a motion control module. The vision processing module is electrically connected to the industrial camera, and the motion control module is electrically connected to the push rod cylinder. A support and adjustment structure is provided at the bottom of the frame, including multiple adjustable feet for adjusting the levelness of the equipment.

2. The intelligent feeding and sorting device with high-precision visual recognition according to claim 1, characterized in that: The gantry support spans across the horizontal conveyor belt, the industrial camera is positioned at the top center of the gantry support with its lens facing downwards, and the ring light source is arranged around the outer periphery of the industrial camera lens.

3. The intelligent feeding and sorting device with high-precision visual recognition according to claim 2, characterized in that: The visual recognition station also includes a photoelectric sensor, which is located upstream of the industrial camera and is used to detect whether the material has arrived at the visual recognition station and trigger the industrial camera to acquire images.

4. The intelligent feeding and sorting device with high-precision visual recognition according to claim 1, characterized in that: The vision processing module has a built-in image processing algorithm for extracting features and classifying the acquired images, and outputs sorting control signals to the motion control module.

5. The intelligent feeding and sorting device with high-precision visual recognition according to claim 1, characterized in that: Multiple material discharging chutes are arranged side by side along the width of the frame. The inlet end of each material discharging chute corresponds to the end of the horizontal conveyor belt, and the outlet end of each material discharging chute extends to different material collection areas.

6. The intelligent feeding and sorting device with high-precision visual recognition according to claim 5, characterized in that: The sorting execution mechanism also includes a sorting table, which is located between the end of the horizontal conveyor belt and the inlet end of the material dispensing chute, and is used to receive materials falling from the end of the horizontal conveyor belt; the push rod cylinders are respectively located at each sorting station of the sorting table, and are used to push the materials into the corresponding material dispensing chute.

7. The intelligent feeding and sorting device with high-precision visual recognition according to claim 1, characterized in that: The control cabinet is equipped with a human-machine interface for displaying material identification results, sorting statistics and equipment operating status, and for receiving user operation commands.

8. The intelligent feeding and sorting device with high-precision visual recognition according to claim 1, characterized in that: There are four adjustable feet, which are respectively located at the four corners of the bottom of the frame. Each adjustable foot includes a screw, an adjusting nut threadedly connected to the screw, and a base plate fixed to the bottom of the screw.

9. The intelligent feeding and sorting device with high-precision visual recognition according to claim 1, characterized in that: The sorting method of this device includes the following steps: Step S1: The material is guided into the horizontal conveyor belt by the guiding and sorting structure, and passes through the vision recognition station in sequence along the conveying direction; Step S2: When the material arrives at the vision recognition station, the ring light source is lit to provide illumination, and the industrial camera captures the image information of the material and transmits it to the vision processing module of the control cabinet; Step S3: The vision processing module extracts and classifies features from the acquired images to determine the category or quality of the materials and generates corresponding sorting control signals. Step S4: The motion control module controls the corresponding pusher cylinder to move according to the sorting control signal, pushing the material into the corresponding material discharge chute to complete the sorting operation.

Images