CN206139527U - Panoramic vision potato is selected separately and defect detecting device - Google Patents

Abstract

The utility model discloses a panoramic vision potato sorting and defect detection device, which comprises a conveying device, a detection obscura, a sorting mechanism, an infrared sensor module, an image acquisition mechanism, and image processing with a built-in convolutional neural network and a support vector machine (SVM). An analysis module, a data fusion module with a built-in support vector machine (SVM), and a timing module for coordinating the actions of various components. The panoramic vision potato sorting and defect detection device of the utility model can complete all-round detection without turning over the potatoes during the detection process. On the one hand, unnecessary damage to the potatoes is avoided, and on the other hand, the dynamic camera detection The instability of the image is improved, and the accuracy of the detection is improved. The utility model can be widely used for real-time on-line detection of the external quality of potato agricultural products, and has great significance for promoting the development of the potato industry in my country.

Description

Panoramic Vision Potato Sorting and Defect Detection Device

technical field

The utility model relates to a device and a method for detecting the appearance quality of agricultural products, in particular to a potato sorting and defect detection device and a detection method.

Background technique

Shape and surface defects are important characteristics of potato appearance quality. Through quantitative measurement of these characteristic indicators, comprehensive detection and grading of potato external defects, shape and other indicators can be completed.

my country is the largest potato producer in the world, and most of the potato quality inspections are still in the stage of identification and judgment by artificial senses. This method of manual detection and evaluation of potato quality has low efficiency, poor objectivity and accuracy, is difficult to meet the requirements of high-standard grading, and is not conducive to realizing large-scale and automatic quality inspection operations.

The use of machine vision for inspection can eliminate the interference of human subjective factors, and can provide a reliable basis for realizing large-scale and automated quality inspection operations.

At present, the potato grading devices used in most potato processing enterprises generally only classify by weight, use balances or pressure sensors to obtain weight information, and then classify according to the principle of leverage or control circuits. But these devices can only be graded by weight, and cannot automatically pick out defective potato pieces. In the actual application and operation of such equipment, additional manpower is required to pick out the defective products first, and then grade them according to weight, which will increase the cost of potato grading, increase manpower and material resources, thereby increasing production costs, and the grading process cannot really leave Manual participation cannot provide a reliable basis for realizing large-scale and automated quality inspection operations.

Although the research on potato grading methods and equipment based on computer vision has gradually become a hot spot, it is generally limited to laboratory research or the use of a single camera to take pictures of potatoes. There are not many real large-scale applications in actual production and processing. Although some grading algorithms With a relatively high recognition rate, but because only one camera is used, the defect detection of potatoes is not comprehensive, and there is a probability of missed detection.

In order to solve the problems that the existing potato grading equipment can only be graded according to weight or a single camera can not meet the requirements of real-time detection, a panoramic vision potato sorting and defect detection device and method are proposed.

Through the search of domestic patent literature, it is found that there are some relevant patent literature reports, mainly as follows:

1. The patent with the publication number CN 202539096 U discloses a sorting and removing mechanism for fruits and vegetables, especially a sorting and removing mechanism for large fruits and vegetables, which can remove clods, stones and glass mixed in fruits and vegetables. Remove immature fruits. This patent adopts the method of hitting fruits and vegetables head-on, and the sorted fruits and vegetables directly fall into the material bin, which is easy to cause damage to the fruits and vegetables.

2. The announcement number is CN 104056790 A, a utility model patent titled "A Method and Device for Intelligent Potato Sorting", which solves the problem that existing potato grading equipment can only be classified according to weight, and although some equipment can be classified according to appearance characteristics, but The grading algorithm is more complex, and cannot well meet the requirements of real-time detection and other issues.

3. The announcement number is CN 204746897 U, and the name is "a potato grading control device based on machine vision technology". It is a utility model patent, which can realize the rapid detection and sorting of impurities and potatoes of different qualities, and use air jets to remove impurities and control the quality of potatoes The collision angle between the potato and the guide lever is used to reduce the collision force and reduce the mechanical damage of the potato; according to the detection transverse diameter of the potato, one or more guiding mechanisms are used to remove the potato, so as to realize the sorting of the potato to be graded.

4. The notification number is CN 203732461 A, a utility model patent titled "a horizontal conveying and uniform turning device for potato quality image collection", which designs a horizontal conveying and constant speed turning device for potato quality image collection, In the non-destructive testing of potato external quality, it can be turned over smoothly at a uniform speed during horizontal transportation, and the center positioning of the potato testing process can be guaranteed, and the dynamic acquisition of potato external quality images can be realized.

Although the above-mentioned patents propose a potato sorting method and potato grading equipment, although some equipment can be graded according to the appearance characteristics, because only one camera is used, the potatoes need to be turned over, and the defect detection is not comprehensive, and there is a probability of missed detection, and The means of analyzing and processing the photos taken by the camera are relatively simple, and the images cannot be analyzed accurately, quickly and comprehensively, and the analysis results are not satisfactory. In addition, potatoes may be damaged when they are turned over.

Utility model content

The purpose of the utility model is to provide a panoramic vision potato sorting and defect detection device, which can automatically transport potatoes, and can fully photograph and detect potatoes without turning them over, and the detection is more efficient and accurate.

In order to achieve the above object, the panoramic visual potato sorting and defect detection device of the present utility model comprises a conveying device, a detection camera obscura, a sorting mechanism, an infrared sensor module, an image acquisition mechanism, a built-in convolutional neural network and a support vector machine (SVM) image Processing analysis module, data fusion module with built-in support vector machine SVM and timing module for coordinating the actions of various components;

The conveying device includes a frame, on which conveying rollers are arranged at intervals, and each conveying roller is located on the same horizontal plane, and the distance between adjacent conveying rollers is 1-2.5 cm; with the conveying direction as the front, the most forward conveying roller The frame at the front is connected with a blanking plate forward; the frame at the rear end of the conveying roller is connected with a feeding device backward, and the feeding device is set downwards, and the feeding device adopts a belt conveyor or a chain conveyor; The racks on the left and right sides of the rollers are provided with conveying baffle groups for preventing potatoes from falling in the left and right directions; the left end of each conveying roller is provided with transmission gears for transmission connection with the power mechanism;

The rack in the middle of the conveying device is connected upwards with a detection dark box, and the lower part of the front side wall and the lower part of the rear side wall of the detection dark box are correspondingly provided with openings for passing potatoes; the detection dark box is provided with the image acquisition mechanism, upper lighting device and lower lighting. device; the lower lighting device is provided with two;

The two adjacent conveying rollers directly below the middle position in the front and rear direction of the detection box adopt a transparent roller mechanism; the transparent roller mechanism includes a transparent glass roller, a left rubber roller, and a right rubber roller that are hollow and open at both ends , the left supporting device, the right supporting device and the supporting connecting rod; the left and right supporting devices all include a roller bracket and a socket connected to the roller bracket through a rolling bearing; the left end of the socket of the left supporting device is provided with a The transmission gear connected with the transmission of the power mechanism; the left end of the transparent glass roller is inserted into the left rubber roller, the right end of the transparent glass roller is inserted into the right rubber roller, and the left and right rubber rollers are connected to the transparent glass roller. The glass roller is interference fit; the right end of the right rubber roller is plugged into the socket of the right support device and is interference fit with the socket, and the left end of the left rubber roller is plugged into the left support device and interference fit with the socket; the right rubber roller is hollowly set, the right end of the support link is connected to a camera power device, and the camera power device is connected to the right end of the frame, The support link passes through the socket of the right support device and the right rubber roller to the left, and its left end is located in the transparent glass roller;

The gap between the two transparent roller mechanisms forms an infrared sensing channel, and the infrared sensor module includes an infrared emitter and an infrared receiver, and the infrared emitter and the infrared receiver are respectively located on the left and right of the infrared sensing channel square, and both the infrared emitter and the infrared receiver are facing the infrared sensing channel;

The image collection mechanism includes 1 global image collection module for collecting global images of potatoes and 6 local image collection modules for collecting partial images of potatoes; the global image collection modules include a global camera arranged on the top of the rear side wall in the detection obscura;

The six partial image acquisition modules are respectively a rear partial image acquisition module located in the middle of the rear side wall in the detection dark box, a left partial image acquisition module located in the middle of the left wall in the detection dark box, and a left partial image acquisition module located in the middle of the detection dark box. The right partial image acquisition module in the middle of the inner right wall, one front partial image acquisition module located in the middle of the front side wall in the detection obscura, and two lower partial image acquisition modules, the transparent glass of each transparent roller mechanism One of the lower partial image acquisition modules and one of the lower lighting devices are respectively arranged in the roller, and the lower partial image acquisition module and the lower lighting device are both connected to the left end of the support link;

The front, rear, left and right partial image acquisition modules have the same structure, and all include a camera movement mechanism for driving the camera to reciprocate linear motion and a partial camera connected to the camera movement mechanism; each lower partial image acquisition module includes a local camera;

One side of the frame at the exit of the detection dark box is provided with a sorting mechanism for pushing the potatoes away from the conveying device along the left and right directions, and the corresponding conveying baffle group of the sorting mechanism is provided with a gap for passing through the potatoes; the sorting mechanism is Push rod type sorting mechanism or jet type sorting mechanism; the part of the conveying device at the sorting mechanism forms the area to be separated;

The frame in front of the detection camera obscura is connected with an installation frame, and the image processing and analysis module, the data fusion module and the timing module for coordinating the actions of each component are arranged on the installation frame; the timing module is connected with the power mechanism, camera movement The driving device of the mechanism, the infrared sensor module, the sorting mechanism and the image processing and analysis module;

Both the local camera and the global camera are connected to the image processing and analysis module.

The left and right sides of the blanking plate are upwardly connected with blanking baffles. In this way, the potatoes will not fall from the left and right sides when passing through the cutting plate.

The push rod type sorting mechanism includes a sorting power device and a push rod connected with the sorting power device in transmission, and the push rod is provided with a dial for pushing potatoes.

The air-jet sorting mechanism includes an air-jet pipe, one end of which is connected to a high-pressure air source, and the other end is connected with an air nozzle, the air nozzle is located on one side of the frame at the exit of the detection dark box and the mouth of the air nozzle faces the other side of the frame.

The utility model has the following advantages:

The utility model refers to the national export standard, and according to the shape, weight and appearance characteristics of the potato, without turning the potato, the utility model carries out the panoramic visual detection of the potato from top to bottom, left, right, front and back, so as to complete the real-time detection of the potato with grading.

The panoramic vision potato sorting and defect detection device and the sorting detection method of the utility model can complete all-round detection without turning over the potatoes during the detection process, on the one hand avoiding unnecessary damage to the potatoes, on the other hand It avoids the instability in dynamic camera detection, improves image clarity, and improves detection accuracy. The utility model can be widely used for real-time on-line detection of the external quality of potato agricultural products, and has important practical significance and good application prospects for promoting the development of the potato industry in my country.

The panoramic visual potato sorting and defect detection device of the utility model has a reasonable structure, and the steps of the sorting and testing method of the utility model are arranged compactly and efficiently. After the new detection device, the defective potatoes are automatically sorted out to provide support for subsequent processing or sales of potatoes.

Two transparent roller mechanisms are located next to each other so that the captured image completely covers the bottom features of the potato.

Description of drawings

Fig. 1 is the structural representation of the utility model;

Fig. 2 is the flowchart of detection method of the present utility model;

Fig. 3 is the structural representation of detecting dark box;

Fig. 4 is a schematic diagram of an exploded structure of a transparent roller mechanism;

Fig. 5 is a schematic structural diagram of the convolutional neural network built into the image analysis and processing module;

Fig. 6 is a flow chart of data fusion between the image analysis processing module and the data fusion module.

detailed description

The utility model takes the conveying direction of the potatoes as the forward direction; the direction shown by the arrow in Fig. 1 is the conveying direction of the potatoes.

As shown in Figures 1 to 6, the utility model provides a panoramic vision potato sorting and defect detection device, including a conveying device, a detection camera obscura, a sorting mechanism, an infrared sensor module, an image acquisition mechanism, and a built-in convolution neural network. Image processing and analysis module 30 of network and support vector machine SVM, built-in data fusion module 31 of support vector machine SVM and timing module 32 for coordinating the actions of various components;

The conveying device includes a frame 1, and the frame 1 is provided with conveying rollers 2 at intervals, each conveying roller 2 is located on the same level, and the distance between adjacent conveying rollers 2 is 1-2.5 cm, so that the potatoes of normal size do not It will leak from the gap between the adjacent conveying rollers 2; with the conveying direction as the front, the frame 1 at the frontmost conveying roller 2 is connected forward with the blanking plate 3; The frame 1 is connected with a feeding device backwards, and the feeding device is arranged inclined downwards, and the feeding device adopts a belt conveying device or a chain conveying device. When the belt conveyor is adopted, reference numeral 35 in Fig. 1 shows the feeding plate arranged on the conveying belt, and the feeding plate catches the potatoes during work, so that the potatoes are transported to the conveying roller 2 along with the belt. When a chain conveying device is adopted, reference numeral 35 in Fig. 1 shows the upper hopper arranged on the chain, and the upper hopper catches the potatoes during operation, so that the potatoes are transported to the conveying roller 2 places. The left and right sides of the feeding device are provided with feeding baffles 36 for blocking potatoes from falling in the left and right directions.

The frame 1 on the left and right sides of the conveying roller 2 is provided with the conveying baffle group 4 used to stop the potatoes from falling in the left and right directions; The plate group 4) is equipped with transmission gears for transmission connection with the power mechanism; the power mechanism is an ordinary gear transmission mechanism, which is a conventional technology in the field, and its specific structure will not be described in detail, and the figure is not shown.

The frame 1 in the middle part of the conveying device is upwardly connected with a detection box 5, and the lower part of the front side wall and the lower part of the rear side wall of the detection box 5 are correspondingly provided with an opening 6 for passing through potatoes; the detection box 5 is provided with the image acquisition mechanism, upper The lighting device 7 and the lower lighting device 8; (the upper lighting device 7 adopts a ring light, and the lower lighting device 8 adopts an LED lamp), and the lower lighting device 8 is provided with two;

The adjacent two conveying rollers 2 directly below the middle position in the front and rear direction of the detection box 5 adopt a transparent roller mechanism; the transparent roller mechanism includes a transparent glass roller 9, a left rubber roller 10, Right squeegee 11, left support device, right support device and support link 15; Left and right support device all comprise roller bracket 12 and the plug socket 14 that is connected on the roller bracket 12 by rolling bearing 13; Left support device The left end of the plug-in barrel 14 is provided with a transmission gear for transmission connection with the power mechanism (the transmission gear is a conventional structure, not shown in the figure); the left end of the transparent glass roller 9 is inserted into the left rubber roller 10, and the transparent glass The right end of the roller 9 is inserted into the right squeegee 11, and the left and right squeegee 10, 11 are interference fit with the transparent glass roller 9; the right end of the right squeegee 11 is inserted into the right The inserting sleeve 14 of the supporting device and interference fit with the inserting sleeve 14, the left end of the left rubber roller 10 is inserted in the inserting sleeve 14 of the left supporting device and is interference fit with the inserting sleeve 14 The right squeegee 11 is hollowly set, and the right end of the support link 15 is connected to a camera power device in transmission, and the camera power device is connected to the right end of the frame 1 . The camera power device adopts various common forms such as air cylinders, hydraulic cylinders, and electric push rods, which are conventional technologies in the art, and are not shown in the figure. During work, under the control of the timing module 32, the camera power unit drives the local camera of the local image acquisition module below to move to a position suitable for taking pictures under the potato through the support connecting rod 15.

The support connecting rod 15 passes through the inserting cylinder 14 and the right rubber roller 11 of the right supporting device to the left and its left end is located in the transparent glass roller 9;

The gap between the two transparent roller mechanisms forms an infrared sensing channel 16, and the infrared sensor module includes an infrared transmitter 17 and an infrared receiver 18, and the infrared transmitter 17 and the infrared receiver 18 are located in the infrared sensing channel respectively. 16 to the left and right, and the infrared transmitter 17 and the infrared receiver 18 are facing the infrared sensing channel 16;

The image collection mechanism includes 1 global image collection module for collecting global images of potatoes and 6 local image collection modules for collecting partial images of potatoes; 19;

The 6 local image acquisition modules are respectively a rear partial image acquisition module 22 located in the middle of the rear side wall in the detection box 5, and a left partial image acquisition module 23 and 1 located in the middle of the left wall in the detection box 5. A right partial image acquisition module 24 positioned at the middle part of the right side wall in the detection dark box 5, a front partial image acquisition module 25 positioned at the middle part of the front side wall in the detection dark box 5, and two lower partial image acquisition modules 26, each of which The transparent glass roller 9 of the transparent roller mechanism described above is respectively provided with one of the lower partial image acquisition modules 26 and one of the lower lighting devices 8, and the lower partial image acquisition module 26 and the lower lighting device 8 are both connected to the left end of the supporting link 15;

The front, rear, left and right partial image acquisition modules have the same structure, and all include a camera motion mechanism 20 for driving the camera in a reciprocating linear motion and a local camera 21 connected to the camera motion mechanism 20; the camera motion mechanism 20 includes guide rails and the driving device, the camera is slidably connected on the guide rail and connected with the driving device in transmission. The driving device can adopt various common linear driving devices such as cylinder, electric push rod, micro motor and screw mechanism. The camera movement mechanism 20 is a conventional technique in the art, and is not shown in detail in the figure. Each lower partial image acquisition module 26 includes a partial camera 21 .

One side of the frame 1 at the exit of the detection black box 5 is provided with a sorting mechanism for pushing the potatoes away from the conveying device along the left-right direction, and the corresponding conveying baffle group 4 of the sorting mechanism is provided with a gap 27 for passing through the potatoes; The sorting mechanism is a push rod type sorting mechanism or a jet type sorting mechanism; the sorting mechanism shown in Figure 1 is a push rod type sorting mechanism. When a jet-type separation mechanism is used, the jet-type separation mechanism includes a vent pipe, one end of which is connected to an air nozzle, and the other end is connected to a high-pressure cylinder or an air pump. Each component of the push rod type sorting mechanism or the jet type sorting mechanism is a conventional technology in the art, and the specific structure thereof is not shown in detail in the figure.

The part of the conveying device at the sorting mechanism forms the area to be separated.

The frame 1 in front of the detection black box 5 is connected with a mounting frame 29, and the mounting frame 29 is provided with the image processing and analysis module 30, the data fusion module 31 and the timing module 32 for coordinating the actions of each component (in Fig. 1 The image processing and analysis module 30, the data fusion module 31 and the timing module 32 are not specifically shown); the timing module 32 is connected to the power mechanism, the driving device of the camera movement mechanism 20, the infrared sensor module, the sorting mechanism and the image processing and analysis module 30 ;

Both the local camera 21 and the global camera 19 are connected to the image processing and analysis module 30 .

The left and right sides of the blanking plate 3 are upwardly connected with blanking baffles 33 . Thereby make potato can not fall from left and right sides when passing through blanking plate 3.

The sorting mechanism shown in Figure 1 is a push rod type sorting mechanism, and the push rod type sorting mechanism includes a sorting power device (the sorting power device adopts various common forms such as air cylinder, hydraulic cylinder, electric push rod, etc., as shown in Fig. not shown) and a push rod 34 connected to the sorting power unit transmission, the push rod 34 is provided with a dial 28 for pushing potatoes.

The air-jet sorting mechanism includes an air-jet pipe, one end of which is connected to a high-pressure air source (such as an air pump or a compressed air tank), and the other end is connected to an air nozzle, which is located on the side of the frame 1 at the outlet of the detection black box 5 and the air The mouth opening faces the other side of the frame 1. Each component of the jet sorting mechanism is conventional technology, not shown in the figure.

The utility model also discloses a potato sorting and detection method using the above-mentioned panoramic vision potato sorting and defect detection device, which is carried out in turn according to the following steps:

Before starting to sort and detect the potatoes, first use the size and shape regularity of normal non-defective potatoes, and the size, shape regularity and surface defect type information of defective potatoes to carry out the support vector machine SVM of the data fusion module 31 Offline training, construction of support vector machine SVM classifier for online detection;

At the same time, using the eigenvalues corresponding to the Area, Perimeter, and Ellipticity of potatoes under different sizes and shapes in the offline training sample data, the support vector machine SVM of the image analysis and processing module is trained offline to build an online detection The support vector machine SVM classifier; In the process of work, above-mentioned two support vector machine SVMs obtain more and more data, make the method of the present utility model have the characteristic of learning, along with the image of the potato of processing more and more Many, the processing speed and processing accuracy of the utility model all can be improved.

The first step is to manually or mechanically place the potatoes on the feeding device, and then turn on the power mechanism. The feeding plate (or hopper) of the feeding device transfers the potatoes to the conveying roller 2, and the power mechanism drives each conveying roller. Axis 2 rotates to drive the potatoes forward;

The second step is to open the infrared sensor module, and when the potato passes through the infrared sensing channel 16, it blocks the infrared rays sent by the infrared emitter 17; Potatoes are located between two transparent glass rollers 9), and the global camera 19 is controlled to take pictures; the global camera 19 takes pictures of the potatoes and sends the images to the image processing and analysis module 30, and the image processing and analysis module 30 calculates that the potatoes are detected The position in the camera obscura 5 and the position information, shape information and size information of the potatoes are sent to the timing module 32; the timing module 32 controls the camera movement mechanism 20 of the front, rear, left and right partial image acquisition modules and the local image acquisition below The camera power device of the module makes each local camera 21 move to a suitable shooting position in the direction close to the potato, and each local camera 21 (including two cameras located in the transparent glass roller 9) takes pictures of the potato from different directions and then respectively The captured image is sent to the image processing and analysis module 30; in this step, the images collected by each local camera 21 are the original three-channel RGB image, and the pixels of the image are 256*256.

The 3rd step is that the image processing analysis module 30 processes the image received, obtains the surface defect type of potato; The potato position, shape and size information obtained by the image analysis collected by the image acquisition module are sent to the data fusion module 31;

The fourth step is to use the support vector machine SVM classifier constructed by the data fusion module 31 to perform data fusion on the received potato surface defect type information, potato position, shape and size information, judge whether the potato to be tested is qualified, and judge the result Send to timing module 32;

The fifth step is that the timing module 32 controls the power mechanism to start. After the potatoes leave the detection black box 5 forward and arrive at the area to be separated, the timing module 32 controls the sorting mechanism to start, and the unqualified potatoes are passed through the potato on the conveying baffle. The gap 27 in the gap is pushed away from the conveying device, and the qualified potatoes are sent out after being transported to the cutting plate 3 by the conveying device, and the sorting work of potatoes is completed.

In the second step, the image analysis and processing module obtains the shape and size information of potatoes according to the image analysis collected by the global camera 19. The processing process is:

Firstly, the image is binarized, filtered, and shape operations are performed to obtain a binarized image, and the edge detection is performed using the Roberts edge detection operator, and the area, perimeter and Perimeter of the potato area are obtained on the binarized image. Ellipticity; Further, using support vector machine SVM, according to the offline training sample data, the size and shape regularity of potatoes are judged in real time according to the currently detected potato area Area, perimeter Perimeter and ellipticity;

(1) Subtract the current image to be processed and the background image to obtain the foreground pixel part of the potato to be checked; the background image is a picture taken by the global camera 19 when there is no potato;

The conveying condition of potatoes on the conveying device can be accurately detected by the infrared sensor module, thus providing accurate reference signal input for the image processing and analysis module 30 . By subtracting two frames of images, different sets of pixels are obtained.

Grayscale the foreground pixel part of the potato to be inspected to obtain the foreground part;

(2) Extract the edge features; this operation is to obtain the edge features of the potato through the obtained foreground part; specifically, use the Roberts edge detection operator to calculate the foreground part, and obtain a pair of black and white binary images representing the main contour information of the potato;

(3) Global eigenvalue extraction; on the basis of black and white binary images, calculate the Area, Perimeter and Ellipticity of the potato area to be detected;

(4) Appearance and size classification; use the support vector machine SVM of the image analysis and processing module to calculate the shape and size information of the potato based on the calculated area, perimeter and Ellipticity of the potato.

In the third step, the image analysis processing module obtains the processing process of the defect type information of potatoes according to the image analysis collected by the local camera 21 as follows:

(1) Scale the 256*256 original three-channel RGB image collected by the local camera 21 into a 224*224 three-channel RGB image,

Then the zoomed image is identified through the convolutional neural network CNN, which includes 8 layers, the first 5 layers are convolutional layers, and the 6th to 8th layers are fully connected layers. The output 10-dimensional vector represents the probability density distribution that the image belongs to 10 types of potato surface defects. Its convolutional neural network structure CNN is shown in Figure 5, and the data processing flow of the network is as follows:

(2) The input layer of the convolutional neural network is the entire scaled image. As shown in Figure 5, the image is expanded in columns to form 50176 nodes; the nodes of the first layer do not have any connecting lines forward.

(3) Convolute the expanded image to generate three feature extraction maps, then sum, weight, and bias the four pixels in each group in the feature extraction map, and obtain three features through the Sigmoid function map;

(4) Convolve the three generated feature maps again, generate three secondary feature extraction maps after convolution, and then sum and weight the four pixels in each group in the secondary feature extraction maps Value, add bias, and get three secondary feature maps through the Sigmoid function.

(5) Rasterize the three secondary feature maps and connect them into a vector to input to the traditional convolutional neural network to obtain the defect type information of potatoes.

In the potato defect detection of partial images, the classification is completed with the help of convolutional neural network, which is essentially an input-to-output mapping, which can learn a large number of mapping relationships between input and output without any input and output. The exact mathematical expression between them, as long as the convolutional network is trained with known patterns, the network has the ability to map between input and output pairs. The convolutional network is trained with a tutor, so its sample set is composed of vector pairs of the shape: (potato local surface binary input image vector, potato surface defect type output vector). All these vector pairs should come from the actual "running" results of the system to be simulated by the network, and they are collected from the actual running system. Before starting training, all weights are initialized with some different small random numbers.

In the image processing analysis module 30, the offline training sample library of the support vector machine SVM and the offline training sample library of the convolutional neural network CNN can increase the number of samples. In addition, the size, shape regularity and surface defect types of potatoes can be further subdivided with the increase of sample size.

Claims (4)

1. panoramic vision Rhizoma Solani tuber osi sorts and defect detecting device, it is characterised in that：Including conveyer device, detection camera bellows, sorting Mechanism, infrared sensor module, image acquisition mechanism, the image procossing for being built-in with convolutional neural networks and support vector machines Analysis module, the data fusion module for being built-in with support vector machines and the tfi module for coordinating each component actuation；

Conveyer device includes frame, and conveying roller is interval with frame, and each conveying roller is located in same level, adjacent defeated Send between roller bearing at a distance of 1-2.5 centimetre；With conveying direction as front, the frame at conveying roller foremost is connected with forward Blanking plate；Frame at the conveying roller of rearmost end is connected with feeding device, the downward-sloping setting of feeding device, feeding dress backward Put using belt conveyor or chain conveyor；

The frame of the conveying roller left and right sides is provided with for stopping baffle conveying group that Rhizoma Solani tuber osi falls in left-right direction；It is each defeated The left part of roller bearing is sent to be equipped with the travelling gear for being connected with actuating unit；

Frame in the middle part of conveyer device has connected up detection camera bellows, detects that the front side wall bottom of camera bellows is corresponding with rear wall bottom Offer for by the opening of Rhizoma Solani tuber osi；Described image collecting mechanism, upper illuminator and lower illumination are provided with detection camera bellows Device；The lower lighting device is provided with two；

Two neighboring conveying roller immediately below the medium position of detection camera bellows fore-and-aft direction adopts transparent roller mechanism；Transparent rolling Axis mechanism include hollow setting and the clear glass roller bearing of open at both ends, left squeegee, right squeegee, left support device, Right support device and support link；Left and right support meanss include roller support and are connected to roller support by rolling bearing On grafting cylinder；The left part of the grafting cylinder of left support device is provided with the travelling gear for being connected with actuating unit；Thoroughly Left squeegee described in the left end grafting of bright glass roller bearing, right squeegee described in the right-hand member grafting of clear glass roller bearing are left and right Squeegee with the clear glass roller bearing interference fit；The right-hand member of right squeegee is plugged on inserting for the right support device Connect in cylinder and with the grafting cylinder interference fit, the left end of left squeegee be plugged in the grafting cylinder of the left support device and with The grafting cylinder interference fit；The hollow setting of right squeegee, the right-hand member of the support link and a shooting power set transmission connect Connect, image the right part that power set are connected to the frame, support link is to the left through the grafting cylinder of the right support device With right squeegee and its left end be located at the clear glass roller bearing in；

Gap between described two transparent roller mechanisms forms infrared sensing passage, and the infrared sensor module includes infrared Emitter and infrared remote receiver, infrared transmitter and infrared remote receiver are located at the left and right of infrared sensing passage respectively, and Infrared transmitter and infrared remote receiver are just to the infrared sensing passage；

Image acquisition mechanism include 1 for gather Rhizoma Solani tuber osi global image global image acquisition module and 6 be used for gather Topography's acquisition module of Rhizoma Solani tuber osi topography；Global image acquisition module includes being arranged on rear wall top in detection camera bellows The global photographic head in portion；

6 topography's acquisition modules are respectively 1 rear topography in detection camera bellows in the middle part of rear wall and adopt Collection module, 1 be located at the left topography acquisition module in detection camera bellows in the middle part of left side wall, 1 be located at it is right in detection camera bellows Right topography acquisition module, 1 front topography collection being located in detection camera bellows in the middle part of front side wall in the middle part of the wall of side Module and 2 lower section topography acquisition modules, are respectively equipped with 1 in the clear glass roller bearing of the transparent roller mechanism described in each Individual described lower section topography acquisition module and the lower lighting device described in 1, lower section topography acquisition module and lower photograph Bright device is both connected to the left part of the support link；

Front, rear, left and right topography acquisition module structure are identical, include reciprocal straight for driving photographic head to make The cam movement mechanism of line motion and the local cameras being connected in cam movement mechanism；Below each, topography adopts Collection module includes a local cameras；

The frame side in detection camera bellows exit is provided with the sorting mechanism for Rhizoma Solani tuber osi is pushed away conveyer device in left-right direction, Baffle conveying group corresponding to sorting mechanism is provided with for by the breach of Rhizoma Solani tuber osi；The sorting mechanism is push-down separator Structure or jet-propelled sorting mechanism；Conveyer device part at sorting mechanism forms region to be separated；

Installing rack is connected with frame in front of the detection camera bellows, installing rack is provided with described image Treatment Analysis module, number Tfi module according to Fusion Module and for coordinating each component actuation；Tfi module connects the actuating unit, cam movement The driving means of mechanism, infrared sensor module, sorting mechanism and image processing and analyzing module；

The local cameras and global photographic head are all connected with described image Treatment Analysis module.

2. panoramic vision Rhizoma Solani tuber osi according to claim 1 sorts and defect detecting device, it is characterised in that：Blanking plate Left and right sides have connected up discharge flapper, so that Rhizoma Solani tuber osi will not be fallen down from the left and right sides when by blanking plate.

3. panoramic vision Rhizoma Solani tuber osi according to claim 1 and 2 sorts and defect detecting device, it is characterised in that：It is described Push-down sorting mechanism includes the push rod for sorting power set and being connected with sorting power set, and push rod is provided with for pushing away The plate of dynamic Rhizoma Solani tuber osi.

4. panoramic vision Rhizoma Solani tuber osi according to claim 1 and 2 sorts and defect detecting device, it is characterised in that：It is described Jet-propelled sorting mechanism includes air jet pipe, and air jet pipe one end connection high-pressure air source, the other end are connected with valve, and valve is located at detection The opposite side of the frame side in camera bellows exit and valve opening towards frame.