CN118060211A

CN118060211A - Automatic Chinese chestnut sorting system and method

Info

Publication number: CN118060211A
Application number: CN202410366431.3A
Authority: CN
Inventors: 王新浩; 王亚波; 高立刚
Original assignee: Jushi Tangshan Robot Technology Co ltd
Current assignee: Jushi Tangshan Robot Technology Co ltd
Priority date: 2024-03-28
Filing date: 2024-03-28
Publication date: 2024-05-24

Abstract

The invention belongs to the technical field of automatic chestnut sorting, and provides an automatic chestnut sorting system and method, wherein the system comprises the following components: the dispersing and feeding mechanism is used for loading and conveying the Chinese chestnut to be processed, and the dispersing roller disperses the Chinese chestnut loaded on the dispersing roller through self-rotation and disperses the Chinese chestnut to the sorting and conveying channel; the sorting mechanism is used for executing sorting action of the chestnut to be processed; the identification mechanism is electrically connected with the sorting mechanism and is positioned at the front end of the sorting conveyor belt and used for collecting surface images and internal images of the Chinese chestnut to be processed so as to identify the appearance characteristics and the internal characteristics of each Chinese chestnut to be processed, the intelligent sorting module and the control module are arranged on the host computer, the appearance characteristics and the internal characteristics of each Chinese chestnut to be processed can be automatically analyzed, and a blowing and separating command is generated according to the sorting information of the intelligent sorting module so as to control and complete the sorting action of the Chinese chestnut to be processed; and the frame is provided with the dispersion feeding mechanism, the sorting mechanism and the identification mechanism. The invention realizes more accurate chestnut identification and classification sorting treatment, improves the chestnut sorting efficiency, simultaneously reduces the sorting cost and improves the product quality sorting speed.

Description

Automatic Chinese chestnut sorting system and method

Technical Field

The invention relates to the technical field of automatic chestnut sorting, in particular to an automatic chestnut sorting system and method.

Background

At present, the automation degree in the agricultural product sorting field is improved, but the problems of low intelligent degree, low sorting accuracy, low sorting efficiency, undetectable internal decay, large equipment volume, high manufacturing cost and the like also exist in the aspect of Chinese chestnut sorting equipment. Therefore, there is a need for a more efficient, accurate, and intelligent sorting system that reduces sorting costs while improving the efficiency and quality of chestnut sorting.

At present, machine vision-based chestnut sorting equipment exists in China, and the common defects of the equipment are as follows: 1) The manufacturing cost is high. Further, the selling price is high, and small chestnut processing enterprises cannot bear the selling price; 2) The occupied area is large, and a set of sorting equipment is usually more than 10 meters, so that the improvement of the existing workshop is not facilitated for the chestnut processing enterprises; 3) The sorting accuracy is not high, and particularly in the aspect of rot and deterioration detection, the sorting device has no internal rot detection function; 4) The sorting channels are few, the sorting efficiency is low, and the full-automatic process from feeding to sorting cannot be realized.

Accordingly, there is a need to provide an automated chestnut sorting system and method to solve the above-mentioned problems.

Disclosure of Invention

The invention aims to provide an automatic chestnut sorting system and method for solving the problems, so as to solve the technical problems that in the prior art, sorting channels are few, sorting efficiency is low, a full-automatic process from feeding to sorting cannot be realized, and the like.

The first aspect of the present invention provides an automatic chestnut sorting system, comprising: the dispersing and feeding mechanism is used for loading and conveying the Chinese chestnut to be processed and comprises a hopper, a feeding conveyor belt, a dispersing roller, a redundant channel and a circulating conveyor belt, wherein the dispersing roller is a flexible self-rotating device, and the Chinese chestnut loaded on the dispersing roller is dispersed through self-rotation and is dispersed to the sorting conveyor channel; the sorting mechanism is used for executing sorting action of the Chinese chestnut to be processed and comprises a sorting conveyor belt, a position identification device, a pneumatic blowing device and a blanking channel; the identification mechanism is electrically connected with the sorting mechanism and is positioned at the front end of the sorting conveyor belt and used for collecting surface images and internal images of the Chinese chestnut to be processed on the sorting conveyor channel to identify the appearance characteristics and the internal characteristics of each Chinese chestnut to be processed, and the identification mechanism comprises a host, a display, an industrial camera, X-ray imaging equipment, a light supplementing lamp and a lamp box, wherein the host is provided with an intelligent sorting module and a control module, the intelligent sorting module can automatically analyze the appearance characteristics and the internal characteristics of each Chinese chestnut to be processed, and the control module generates a blowing command according to sorting information of the intelligent sorting module so as to control the sorting action of the Chinese chestnut to be processed; and the frame is provided with the dispersion feeding mechanism, the sorting mechanism and the identification mechanism.

According to an alternative embodiment, the dispersion feeding mechanism has at least four dispersion rollers, each having a flexible surface to contact the chestnut to be treated, so that the chestnut to be treated is dispersed evenly by rotation to the sorting conveyor belt of the sorting mechanism.

According to an alternative embodiment, according to the blowing command, the sorting action of the chestnut to be processed is completed jointly by the mutual cooperation of the sorting conveyor belt, the pneumatic blowing device and the blanking channel.

According to an alternative embodiment, the sorting conveyor belt further comprises a friction strip positioned below the sorting conveyor belt, the sorting conveyor belt comprises a chain and carrier rollers, one chain corresponds to one group of gears, a pair of carrier rollers are symmetrically arranged on two sides of one chain along the conveying direction to form two sorting conveying channels, and each pair of carrier rollers is connected by a connecting shaft; the friction strips are fixed on the lower side of the center of the bottom of the sorting conveyor belt, so that when the carrier rollers move along with the sorting conveyor belt, and each carrier roller starts to rotate under the action of friction force generated by each carrier roller and each friction strip when the carrier roller passes through the friction strips, the to-be-processed chestnut located on each carrier roller position on the sorting conveyor channel is driven to rotate along with the rotation of the corresponding carrier roller, and a picture of 360 rotations of the to-be-processed chestnut around the center thereof is taken and obtained.

According to an alternative embodiment each idler has the following shape: the middle part is concave, and two side parts which extend outwards from the middle part to two ends are arranged on the middle part, so that a material carrying position is formed at a gap part between two adjacent carrier rollers in the conveying direction of the sorting conveyor belt and is used for carrying each chestnut to be processed.

According to an alternative embodiment, the redundant channel is located below the sorting conveyor belt and is used for collecting scattered Chinese chestnut to be processed; the first end of the circulating conveyor belt is connected with the redundant channel, the other end of the circulating conveyor belt is connected with the feeding conveyor belt, and the circulating conveyor belt is used for conveying the Chinese chestnut to be processed collected from the redundant channel back to the hopper.

According to an alternative embodiment, a plurality of discharge channels are installed at the lower end of the frame, the plurality of discharge channels being equally spaced apart and evenly distributed, and both ends of each discharge channel being located outside both ends of the frame.

According to an alternative embodiment, the frame comprises three partial frames corresponding to the scattered feeding mechanism, the sorting mechanism and the identification mechanism, wherein the three partial frames are mutually connected and integrally formed; or the three partial frames are independent frames and are mutually connected.

The second aspect of the present invention provides a method for automatically sorting Chinese chestnut, which uses the automatic sorting system for Chinese chestnut according to the first aspect of the present invention to automatically sort Chinese chestnut, comprising: the Chinese chestnut to be processed in the hopper is transmitted to a dispersing roller of a dispersing and feeding mechanism, and the Chinese chestnut to be processed is dispersed by the self-rotation of the dispersing roller and is dispersed to a sorting and conveying channel; collecting surface images and internal images of the Chinese chestnut to be processed on the sorting and conveying channel, and identifying appearance features and internal features of the Chinese chestnut to be processed by adopting a pre-trained Chinese chestnut identification model to determine identification and classification results, wherein the method specifically comprises the following steps of: optimizing the target detection frame position loss, confidence loss, category loss and angle regression loss by improving a loss function so as to optimize the chestnut recognition model;

and determining a sorting strategy according to the identification and sorting result, and blowing the chestnut to be processed out of the sorting conveying channel by using a blowing and separating device so as to fall into a corresponding blanking channel to finish sorting operation.

The embodiment of the invention has the following advantages:

Compared with the prior art, the Chinese chestnut sorting machine has the advantages that the Chinese chestnut sorting machine automatically loads the Chinese chestnut through the dispersing and loading mechanism, the Chinese chestnut loaded on the dispersing roller is dispersed through the self-rotation of the dispersing roller of the sorting mechanism and is dispersed to the sorting conveying channel, the appearance image, the internal image and the position information of the Chinese chestnut are collected through the identifying mechanism, the size and the quality of the Chinese chestnut to be processed can be accurately identified, more accurate Chinese chestnut identification and sorting treatment are realized, sorting cost can be reduced while sorting efficiency of the Chinese chestnut is improved, and the sorting speed of product quality can be improved.

In addition, the friction strips are additionally arranged, so that friction force generated by the carrier roller and the friction strips when the carrier roller passes through the friction strips can be utilized to drive the Chinese chestnut to be processed on the sorting conveying channel to rotate along with rotation of the carrier roller, and the problem of material overlapping can be effectively avoided when the Chinese chestnut to be processed is effectively sorted and conveyed to a material carrying position.

Drawings

Fig. 1 is a schematic perspective view of an example of an automatic chestnut sorting system according to the present invention;

fig. 2 is a perspective view illustrating an angle of the automatic chestnut sorting system of fig. 1;

Fig. 3 is a schematic perspective view of another angle of the automatic chestnut sorting system of fig. 1;

fig. 4 is a schematic view of a partial structure of the automatic chestnut sorting system of fig. 1;

Fig. 5 is another partial schematic view of the automatic chestnut sorting system of fig. 1;

fig. 6 is a schematic view of a further partial structure of the automatic chestnut sorting system of fig. 1;

Fig. 7 is a schematic flow chart of an example of the automatic chestnut sorting method of the present invention;

Fig. 8 is a schematic diagram of a network structure of a chestnut recognition model in the automatic chestnut sorting method of the present invention.

Detailed Description

It should be noted that, without conflict, the embodiments of the present application and features of the embodiments may be combined with each other. In the present application, the upper surface of the object in the drawing is taken as the upper surface, and the lower surface of the object in the drawing is taken as the lower surface. Further, spatially relative terms, such as "above … …," "above … …," "on the upper surface of … …," "above," and the like, may be used herein for ease of description to describe one device (or apparatus) or feature's spatial location relative to another device or feature as illustrated in the figures. It will be understood that the spatially relative terms are intended to encompass different orientations in use or operation in addition to the orientation of the device or apparatus depicted in the figures. For example, if the device or apparatus in the figures is turned over, elements described as "above" or "over" other devices or structures would then be oriented "below" or "beneath" the other devices or structures. Thus, the exemplary term "above … …" may include both orientations "above … …" and "below … …". May be positioned in other different ways, such as rotated 90 degrees or at other orientations, and the spatially relative descriptors used herein interpreted accordingly. In this context, the upper side of the drawing sheet is taken as the upper side, the lower side of the drawing sheet (i.e., the side opposite to the upper side) is taken as the lower side, the left side of the drawing sheet is taken as the left side, the right side of the drawing sheet (the side opposite to the left side) is taken as the right side, the horizontal direction of the drawing sheet is taken as the horizontal direction, and the vertical direction of the drawing sheet (the direction perpendicular to the horizontal direction) is taken as the vertical direction. The foregoing is merely illustrative of the apparatus or device for the sake of clarity and is not intended to limit the application thereto.

In order to solve the above problems, the present invention combines machine vision and artificial intelligence techniques. Through simpler structure, realize going up from the chinese chestnut to the automated system of letter sorting to cooperate and use industry camera, X ray imaging equipment and advanced image processing algorithm to realize the discernment to chinese chestnut more accurate and classify letter sorting processing, when improving chinese chestnut letter sorting efficiency, can reduce letter sorting cost, can improve the speed of product quality classification, and then brought essential technological progress for the agricultural product letter sorting field. The automatic chestnut sorting system of the present invention will be described in detail with reference to the accompanying drawings in combination with embodiments.

Example 1

Referring to fig. 1,2, 3,4, 5 and 6, a first aspect of the present invention provides an automatic chestnut sorting system.

Fig. 1 is a perspective view schematically showing an example of the automatic chestnut sorting system of the present invention. Fig. 2 is a perspective view schematically illustrating an angle of the automatic chestnut sorting system of fig. 1.

As shown in fig. 1 and 2, the automatic chestnut sorting system 1000 of the present invention includes a frame 100, a dispersion feeding mechanism 200 mounted thereon, a sorting mechanism 300 and an identification mechanism 400, wherein the dispersion feeding mechanism 200 is used for loading and conveying the chestnut to be processed, the sorting mechanism 300 is used for executing the sorting action of the chestnut to be processed, and is electrically connected with the identification mechanism 400 for executing the sorting action of the chestnut to be processed in cooperation with the sorting mechanism 300.

In a specific embodiment, the rack 100 includes three partial frames corresponding to the discrete feeding mechanism 200, the sorting mechanism 300, and the identifying mechanism 400, where the three partial frames are independent frames and are connected to each other. But is not limited thereto, in other embodiments, the three partial frames are connected to each other and integrally formed. The foregoing is illustrative only and is not to be construed as limiting the invention.

Specifically, the dispersion feeding mechanism 200 includes a hopper 210, a feeding conveyor 220, a dispersion roller 230, a redundant channel 240, and a circulation conveyor 250, wherein the dispersion roller 230 is a flexible self-rotating device, and the chestnuts placed on the dispersion roller 230 are dispersed by self-rotation and dispersed to the sorting conveyor channel.

Fig. 3 is a schematic perspective view of another angle of the automatic chestnut sorting system of fig. 1.

As can be seen in fig. 1,2 and 3, the sorting mechanism 300 includes a sorting conveyor 310, a position recognition device 320, a pneumatic blow-off device 330 (specifically via a blow nozzle 331 as shown in fig. 5), and a discharge chute 340.

Fig. 4 is a schematic view of a partial structure of the automatic chestnut sorting system of fig. 1. Fig. 5 is another partial schematic view of the automatic chestnut sorting system of fig. 1.

As shown in fig. 4 and 5, the sorting conveyor 310 includes several pairs of side plates 352 above, and the position recognition devices 320 (only one position recognition device is shown in fig. 5, but in practice, one position recognition device is associated with each sorting conveyor 350) are provided at positions of each side plate 352 corresponding to the sorting conveyor 350. For example, the position recognition device 320 is installed right in front of a gear (e.g., a driving gear) of any one of the sorting conveyor channels 350 (corresponding to the right side in fig. 4), and the position recognition device 320 calculates a moving distance of the chestnut to be processed by recognizing the number of idlers of each sorting conveyor channel 350.

Alternatively, the dispersion feeding mechanism 200 has at least four dispersion rollers, each having a flexible surface to contact the chestnut to be treated, so that the chestnut to be treated is dispersed uniformly by rotation to the sorting conveyor of the sorting mechanism.

Fig. 6 is a schematic view of a further partial structure of the automatic chestnut sorting system of fig. 1.

As shown in fig. 6, the sorting conveyor 310 includes a chain 353 and carrier rollers 351, one chain 353 corresponds to a set of gears (see fig. 5 and 6, which each show only one side of the gears 356), and a pair of carrier rollers 351 are symmetrically disposed on both sides of the chain 353 along the conveying direction to form two sorting conveying paths 350, and each pair of carrier rollers 351 is connected by a connecting shaft 354.

In this example, the number of sorting conveyor channels 350 is 12, preferably 4 to 16, more preferably 12. Through the increase of the number of letter sorting conveying passageway setting, can effectively improve the separation volume of unit time, can promote separation efficiency promptly.

As shown in fig. 2, the feeding conveyor 220 is inclined from the lower left side to the upper right side, and conveys the chestnut to be processed in the hopper 210 obliquely from the lower to the upper side, and to the dispersing roller 230 of the dispersing and feeding mechanism 200. In the example of fig. 2, the dispersion feeding mechanism 200 has four dispersion rollers 230.

The number of the dispersing rollers 230 of the dispersing and feeding mechanism 200 is not particularly limited, and the above description is merely given as an alternative example, and the present invention is not limited thereto.

The recognition mechanism of the automatic chestnut sorting system of the present invention will be described below.

Specifically, the recognition mechanism 400 is electrically connected to the sorting mechanism 300, and is located on the upstream side in the conveying direction of the sorting conveyor 310 (conveying direction from the hopper to the sorting drive path). The recognition mechanism 400 is used for collecting the surface image and the internal image of the chestnut to be processed on the sorting and conveying channel 350 to recognize the appearance characteristic and the internal characteristic of each chestnut to be processed.

In the example of fig. 1, the identifying mechanism 400 includes a host (including a display and a control module), an industrial camera 420, an X-ray imaging device 430, a light supplementing lamp 440, and a light box 450, wherein the host 410 is, for example, a computer including a control module (including a control program or control software), the host 410 is installed with an intelligent sorting module and a control module, the intelligent sorting module can automatically analyze appearance characteristics and internal characteristics of each chestnut to be processed to obtain sorting information, and the control module generates a blowing command according to the sorting information obtained by the intelligent sorting module to control and complete sorting actions of the chestnut to be processed. The industrial camera 420 is used for collecting appearance images of the chestnut to be processed on the sorting conveyor 310 (in particular on the sorting conveyor 350), and the X-ray imaging device 430 is used for collecting internal images of the chestnut to be processed on the sorting conveyor 310 (in particular on the sorting conveyor 350) so as to identify internal features and appearance features such as size, wormholes, cracks, surface mildew and the like of each chestnut to be processed.

Further, the identifying mechanism 400 further includes a position identifying device, such as a position sensor, an infrared detector, etc. The position identifying device is used for detecting position information of the chestnut to be processed on the sorting conveyor 310 (in particular on the sorting conveyor 350), for example, the relative position of each chestnut to be processed relative to the center point of the sorting conveyor 310, etc.

In the invention, the material loading mode of the sorting conveyor belt is carrier roller loading, and the carrier rollers are matched to form a one-by-one material loading position, but if the carrier rollers cannot rotate, the problem of overlapping of materials caused by material irregularity exists. In order to solve the above-described problems, the present invention improves the structure of the sorting conveyor portion of the sorting mechanism 300, as will be described in detail below with reference to fig. 4.

As shown in fig. 4, the sorting mechanism 300 further includes a friction bar 370 located below the sorting conveyor 310, where the friction bar 370 is fixed on the bottom center lower side of the sorting conveyor 310 (specifically, the bottom center lower side of the connecting shaft 354 of the carrier roller), so that when the carrier rollers 351 follow the chain motion of the sorting conveyor 310, and each carrier roller 351 starts to rotate under the friction force generated by each carrier roller 351 and the friction bar 370 when the carrier roller passes through the friction bar 370, the to-be-processed chestnut located at each carrier roller 351 on the sorting conveyor channel rotates along with the rotation of the corresponding carrier roller 351, so as to take a picture of the to-be-processed chestnut rotating 360 around its central axis, and place one to-be-processed chestnut at each carrier roller position on the sorting conveyor channel, which can effectively solve the problem of overlapping of materials caused by the irregularity of the materials.

Through addding friction strip 370, can utilize bearing roller 351 when passing through friction strip 370 with the produced frictional force of friction strip 370 drives be located on the letter sorting conveying passageway wait to handle the chinese chestnut and follow the rotation of bearing roller 351 and rotate, can effectively avoid the problem that the material overlaps when effectively sorting conveying to carrying the material position.

In addition, for overlapped material (specifically chestnut), it is recovered through redundant channel 240 and is continuously transferred to hopper 210.

Specifically, the carrier roller 351 is a rotatable roller, and is disposed perpendicular to the side plate 351. More specifically, a plurality of idlers 351 may be linked together and rotated following the conveying movement of the sorting conveyor 310.

Specifically, the industrial camera 420, the X-ray imaging device 430 and the position recognition device 320 are electrically connected to the control module of the host 410, so that the appearance image collected by the industrial camera 420, the internal image collected by the X-ray imaging device 430 and the chestnut position information detected and obtained by the position recognition device are sent to the control module of the host, and the intelligent sorting module of the host automatically analyzes the appearance characteristic and the internal characteristic of each chestnut to be processed based on the received appearance image, internal image and position information to obtain sorting information.

Specifically, according to the blowing command, the sorting conveyor 310, the pneumatic blowing device and the discharging channel cooperate with each other to jointly complete the sorting action of the chestnut to be processed.

More specifically, the control module generates a blow-off command according to the analysis information, and controls the blow nozzle 331 (see fig. 5 in detail) of the pneumatic blow-off device 330 to blow off the chestnut to be processed to the corresponding sorting conveying passage 350 according to the blow-off command.

Referring to fig. 3 and 5, the redundant path 240 is located below the sorting conveyor 310 for collecting scattered chestnuts to be processed.

Specifically, a first end of the endless conveyor belt 250 is connected to the redundant path 240, and the other end of the endless conveyor belt 250 is connected to the feeding conveyor belt 220, and the redundant path 240 is used for collecting the chestnut to be processed falling from the process of being transferred from the dispersing roller 230 to the sorting conveyor path 350.

As shown in fig. 3, a plurality of lower channels 340 are installed at the lower end of the frame 100, the plurality of lower channels 340 being uniformly distributed at equal intervals, and both ends of each lower channel 340 being located outside both ends of the frame, specifically, located on the upper side than the upper end of the frame 100, and located on the lower side than the lower end of the frame 100.

Further, the circulating conveyor 250 is used for conveying the chestnut to be processed collected from the redundant channel 240 and the blanking channel back to the hopper 210.

It should be noted that the foregoing is merely illustrative of the present invention and is not to be construed as limiting thereof.

Example 2

The following is an example of a method according to the invention, the system of the first aspect of the invention being particularly suitable for use in the method according to the invention. For details not disclosed in the method embodiment of the present invention, please refer to the system embodiment of the present invention.

Referring to fig. 1, 7 and 8, the automatic chestnut sorting method of the present invention performs automatic chestnut sorting by using the automatic chestnut sorting system of embodiment 1 of the present invention.

The automatic sorting method for the Chinese chestnut comprises the following steps.

Step S101: the Chinese chestnut to be processed in the hopper is transmitted to a dispersing roller of a dispersing and feeding mechanism, and the Chinese chestnut to be processed is dispersed through self-rotation of the dispersing roller and is dispersed to a sorting and conveying channel.

Step S102: collecting surface images and internal images of the Chinese chestnut to be processed on the sorting and conveying channel, and identifying appearance features and internal features of the Chinese chestnut to be processed by adopting a pre-trained Chinese chestnut identification model to determine identification and classification results, wherein the method specifically comprises the following steps of: optimizing the chestnut recognition model by improving a loss function to optimize target detection frame position loss, confidence loss, category loss, and angle regression loss.

Step S103: and determining a sorting strategy according to the identification and sorting result, and blowing the chestnut to be processed out of the sorting conveying channel by using a blowing and separating device so as to fall into a corresponding blanking channel to finish sorting operation.

In one embodiment, when the automatic sorting process of the chestnut is started, the feeding conveyor 220 for conveying the chestnut to be processed in the hopper 210 to the dispersion feeding mechanism 200 is controlled. Next, the feeding conveyor 220 loaded with the chestnut to be treated is conveyed to the dispersion roller 230 by the dispersion feeding mechanism 200. When the automated sorting process of the chestnut starts, the plurality of dispersing rollers 230 (for example, four) start to self-rotate, so that the chestnut to be processed transferred to the dispersing rollers 230 is dispersed to the plurality of sorting transfer paths.

Through the increase of the number of letter sorting conveying passageway setting, can effectively improve the separation volume of unit time, can promote separation efficiency promptly.

Preferably, the friction strips 370 below the sorting conveyor 310 (the friction strips 370 are fixed on the lower side of the bottom center of the sorting conveyor 310, specifically, the lower side of the bottom center of the connecting shaft 354 of the carrier roller), so that when the carrier roller 351 moves along with the chain of the sorting conveyor 310 and each carrier roller 351 passes through the friction strips 370 at the carrying position, the friction force generated by each carrier roller 351 and the friction strips 370 starts to rotate and drives the chestnut to be processed at each carrying position on the sorting conveyor channel to rotate along with the rotation of the corresponding carrier roller 351, each carrying position on the sorting conveyor channel can carry one chestnut to be processed, and the problem of overlapping of materials caused by the irregularity of the materials can be effectively solved.

Optionally, sorting and conveying conditions are monitored in real time, position information of the chestnut to be processed on the sorting and conveying channels is identified, and appearance images and internal images of the chestnut are collected. Specifically, the chestnut to be processed on each sorting and conveying channel is monitored in real time. Specifically, whether the device falls in the conveying process or not and whether the device shoots in a shooting area or not are included.

According to the collected appearance image and internal image of chestnut, adopting a pre-trained chestnut identification model to identify the appearance characteristic and internal characteristic of chestnut to be processed so as to determine the identification and classification result, specifically comprising: optimizing the chestnut recognition model by improving a loss function to optimize target detection frame position loss, confidence loss, category loss, and angle regression loss.

The chestnut recognition model is built by adopting yolov-obb original network architecture, wherein a main network adopts CSPDARKNET53 network structure, multi-scale feature fusion is introduced into the neck part of YOLOv-obb, and an improvement of a P2 small target detection layer is added, so that the model can be more flexible and accurate in processing targets with different scales. The network structure of the improved chestnut recognition model is shown in fig. 8.

Specifically, in CSPDARKNET, a CSP connection structure is introduced that allows the feature maps of the different stages to be interchanged for better capture of chestnut features.

In the application scene of object recognition (in particular, in the application scene of recognizing chestnut), feature images and features generally refer to representative image information extracted from an image to be processed. In the deep learning model, the image data to be processed (i.e. the original image data) can be converted into a feature map through a Convolutional Neural Network (CNN) and other structures, wherein each pixel point represents the feature of the corresponding position in the image to be processed (specifically, the chestnut appearance image and the chestnut internal image).

In object recognition, feature maps are very important because they contain abstract feature information at different locations in the image, which is critical for object recognition and localization. Features generally refer to representative feature vectors or feature descriptors extracted from a feature map, and are typically used to represent semantic information for different regions in an image. In the present invention, the target comprises wormholes, tears, surface mold, internal rot.

Specifically, CSP connection divides the feature map into two parts, applies some operations between the two parts, and then connects them together. This way of attachment helps to improve the reusability of features and the flowability of information. At some stage of the network, a max pooling layer is typically used to reduce the size of the feature map in order to reduce the computational complexity and increase the receptive field of the network.

In order to perform more efficient sorting classification treatment (such as efficient defect classification and size classification, and reduce sorting classification time consumption) on the Chinese chestnut to be processed, the invention improves a backbone network and selects a MobileNetV network structure with lighter weight. Specifically, mobileNetV backbone networks employ depth separable convolution. This convolution structure integrates the standard volume into two stages of deep convolution and point-by-point convolution, thereby significantly reducing the number of parameters and the amount of computation. The depth separable convolution can reduce the computational cost while maintaining good feature representation capabilities.

MobileNetV3 is used instead of CSPDARKNET53.

For the channel attention mechanism, for example, a characteristic image X of the appearance image of the chestnut to be processed is input, wherein the size of the characteristic image X is C multiplied by H multiplied by W, C represents the number of channels, H represents the height, and W represents the width. The weight of each channel is calculated by the channel attention mechanism to represent its importance in the overall feature map.

In deep learning, a channel refers to each channel in the feature map, and represents different feature information in the feature map. For example, in the characteristic map of the chestnut appearance image, each channel may correspond to a different color, texture, shape, etc. Through the channel attention mechanism, the weight of each channel can be calculated to represent its importance throughout the feature map.

In the application scene, the whole feature map corresponds to the feature map extracted from the chestnut appearance image. These feature maps contain feature information, such as color, texture, shape, etc., for different locations and scales in the image. Through a channel attention mechanism, each channel in the whole feature map is weighted to extract the most important feature information, and the performance and accuracy of the model in the chestnut recognition task can be improved.

Specifically, global information of the channel is obtained by global average pooling of the feature map X:

Global Pooling(X)＝AvgPool(X)

Wherein Global Pooling (X) is used for carrying out global average pooling on a characteristic graph X of the Chinese chestnut appearance image to obtain global information of the channel; avgPool (X) represents performing a global average pooling operation on the feature map X of the chestnut-appearance image. The global information of each channel is obtained in the step; x represents the characteristic diagram of the Chinese chestnut appearance image.

The globally pooled feature map X is then input into a fully connected layer (or a convolutional layer) and an activation function (e.g., reLU) is used to derive channel attention weights:

Attention_Weights＝σ(FC(Global Pooling(X)))

Wherein, attention_ Weights represents the channel Attention weight; FC (Global Pooling ()) represents the fully connected layer and σ represents the activation function.

Finally, multiplying the channel attention weight with the feature map X to obtain a final feature representation in a weighted summation manner: output=attention_ Weights +.X

Wherein Output represents the feature map weighted by the channel attention mechanism; the addition of the elements indicates the multiplication operation; attention_ Weights represents the channel Attention weight; x represents a characteristic diagram of a chestnut appearance image of the chestnut to be treated. Thus, the important channels are weighted higher, thereby improving the characterizations of, for example, appearance features and interior features.

By introducing a channel attention mechanism, the attention can be dynamically adjusted according to the importance of different channels in the feature map, so that the network focuses more on important feature channels, and the characterization capability and detection performance of the features are improved. The attention mechanism can help the model to better understand the relation between different channels in the feature map in the target detection task, so that the accuracy and the stability of detection are improved.

Further, introducing multiscale feature fusion and adding an improvement of a P2 small target detection layer in the neg part YOLOv-obb can enable the model to be more flexible and accurate in processing targets with different scales. The following is a detailed description of these two improvements.

For multi-scale feature fusion, the multi-scale feature fusion aims at fusing feature graphs from different levels so as to improve the perceptibility of the model to different scale targets. Multi-scale feature fusion may be achieved by the following steps.

Step S201: and extracting the characteristics of the marked images in the training data set.

Specifically, feature maps of different levels are obtained from the backbone network, such as P3 (e.g., a feature map of 80 pixels×80 pixels), P4 (e.g., a feature map of 40 pixels×40 pixels), P (e.g., a feature map of 20 pixels×20 pixels), and the like.

The training data set comprises chestnut appearance images marked with sizes, wormholes, cracks and surface mildews.

Step S202: and (5) performing scale adjustment.

For each feature map, the up-sampling or down-sampling method is used to scale it so that all feature maps have similar resolution. The feature diagrams after passing through the resizing process are, for example, F _p3、F_p4 and F _p5.

Step S203: and fusing the adjusted characteristic diagrams. For example, the following fusion method is adopted, and specifically includes cascade fusion, element weighted fusion and the like. Assume that the fused feature map is F _fusion.

The following expression is used to represent the process of feature fusion:

F_fusion＝α₃×F_p3+α₄×F_p4+α₅×F_p5

Wherein, alpha 3, alpha 4 and alpha 5 are weight parameters for adjusting each scale feature map, and the sum is 1; f _p3、F_p4、F_p5 represents the resized feature maps corresponding to the feature maps P3 (for example, a feature map of 80 pixels×80 pixels), P4 (for example, a feature map of 40 pixels×40 pixels), and P5 (for example, a feature map of 20 pixels×20 pixels) of the different levels, respectively.

Targets of different dimensions are involved for the target detection task, for example, targets of large, medium and small dimensions are detected simultaneously. In this case, the model may be made more focused on features at different scales by adjusting the weight parameters of each scale feature map.

Specifically, the application scenario of the present invention requires that the detected targets be distributed on different scales, as follows: large scale target: such as an object of equivalent size to a vehicle, etc.; mesoscale target: such as an object of equivalent size to a person, etc.; small scale target: such as an object of equivalent size to a small animal, etc.

According to the application scene, the weight parameters of the multi-scale feature fusion can be adjusted, so that the model can better perceive different targets under different scales.

Specifically, α3 corresponds to a feature map containing large scale target features, such as the P3 layer. This feature map may provide more semantic information about large scale objects, such as objects of the same size as vehicles, etc. α4 corresponds to a feature map containing mesoscale target features, such as the P4 layer. This feature map may provide more semantic information about the mesoscale target, such as an object of equivalent size to a person, etc. α5 corresponds to a feature map containing small scale target features, such as the P5 layer. This feature map may provide more semantic information about small scale objects, such as objects of equivalent size to small animals, etc.

Therefore, the weight parameter of each scale feature map is adjusted according to the importance of different scale targets, so that the model can better sense and detect the targets with different scales, and further the model is more accurate and robust in the target detection task.

Preferably, the method further comprises adding a P2 small target detection layer.

And classifying the chestnut to be processed under the image visual field of the equipment, and classifying by adopting the appearance characteristics of the chestnut to be processed, such as the chestnut with cracks (namely cracks, particularly cracked long-strip toothed characteristics), the chestnut with mold (namely the surface of the chestnut is moldy, particularly the appearance of the chestnut is covered with white bluish mold characteristics), and the chestnut with wormholes (particularly the appearance surface of the chestnut is provided with circular inward concave characteristics), wherein the characteristics belong to small targets under the visual field, and a P2 small target detection layer is added on the original structure. The purpose of adding the P2 small target detection layer is to specially process the small target so as to improve the detection precision of the small target. The addition of the P2 small target detection layer can be achieved by:

first, a feature map Fp2 of the P2 layer is acquired from the backbone network.

Then, an additional object detection layer is added on Fp2 for detecting small-sized objects. This target detection layer may take a similar structure as the other detection layers in YOLOv, for example using a convolution operation to generate the detection result.

Specifically, assuming that the output of the P2 small target detection layer is Dp2, the output of the P2 small target detection layer can be expressed by the following formula:

D_p2＝Conv(F_P2)

Wherein Dp2 represents the output of the P2 small target detection layer (e.g., a 160 pixel by 160 pixel feature map), specifically a feature map, which contains the result of the convolution operation of the P2 layer, and can be used to detect a small-sized target; conv (F _P2) represents convolution operation, specifically, an output characteristic diagram generated by carrying out convolution operation on an input characteristic diagram F _P2; fp2 represents a profile of the P2 layer, which is one layer in the backbone network (backbone). In YOLOv, the P2 layer refers to a feature map with a certain downsampling rate, typically used to detect smaller size targets.

By introducing multi-scale feature fusion and adding a P2 small target detection layer, the model can be more flexible and accurate in a target detection task. The multi-scale feature fusion can provide richer feature representation, and the perceptibility of the model to targets with different scales is improved. By adding the P2 small target detection layer, small targets such as insect eyes can be handled exclusively, and the detection accuracy of the small targets such as insect eyes can be improved.

After the network structure reconstruction of the chestnut identification model is completed, a training data set is established, wherein the training data set comprises chestnut appearance images marked with the following category labels: whether the chestnut is in the specified size range, whether the chestnut has cracks, whether the chestnut has surface mildewing and whether the chestnut has wormholes. Also comprises a Chinese chestnut internal image marked with whether the Chinese chestnut is internally rotten or not.

In this example, the single target detection frame is changed to the multi-target detection frame, and the storage mode of the sample data in the training data set, that is, the storage of the appearance image of the chestnut and the internal image of the chestnut to be processed is to store the position information of the rotation frame containing the target. The training data set is for example an OBB data set.

It should be noted that, since the split is relatively common, the split occupies about 60% of the training data set, and is not common like wormholes, surface mildew and internal rot, the training data set has serious unbalance in the category of the sample data, and the rotation angle is added.

In order to solve the problems, the invention improves the Loss function of the chestnut recognition model and introduces Focal Loss to help solve the problem of unbalanced category in target detection.

It should be noted that Focal Loss is a weighted Loss function for difficult-to-classify samples (e.g., a small number of samples or samples with unbalanced positive and negative samples), which can make the chestnut recognition model pay more attention to the difficult-to-classify image samples, so as to improve the learning ability of the model for the difficult-to-classify samples.

Specifically, optimizing the target detection frame position loss, confidence loss, category loss, angle regression loss by improving a loss function to optimize the chestnut recognition model comprises:

L＝L_coord+L_conf+L_cls+λL_angle

Wherein, L represents the total loss function of the chestnut recognition model, and specifically comprises confidence loss, category loss and angle regression loss of the appearance image of the chestnut in the training data set, the position loss of the target detection frame in the internal image of the chestnut, whether the target exists in the target detection frame and the accuracy of the existing target; λ is the weight of the control angle regression loss in the total loss; l _coord is used for calculating the loss values of the center coordinates, width and height of a target detection frame in the appearance image and the internal image of the Chinese chestnut by using the smooth L1 loss; l _conf represents whether a target exists in the target detection frame and the confidence loss of the accuracy of the existence of the target, wherein the target comprises insect eyes, cracks, surface mildew and internal corrosion; l _cls represents the loss value of one or more target categories in the appearance image of the Chinese chestnut in the training data set and the loss value of the target category in the internal image of the Chinese chestnut by using cross entropy loss calculation; l _angle represents an angle regression loss value of a rotation angle of a target detection frame in the Chinese chestnut appearance image in the training data set relative to an image coordinate system and an angle regression loss value of a rotation angle of a target detection frame in the Chinese chestnut inner image in the training data set relative to the image coordinate system.

The boundary box coordinate loss (Bounding Box Coordinate Loss) is calculated by using the smoothed L1 loss, and the loss of the center coordinates x, y and the dimensions w, h is denoted as L _coord.

The confidence loss L _conf of whether a target exists in the target detection frame and the accuracy of the existence of the target is calculated by adopting the following expression:

Wherein L _conf represents whether a target is present within the target detection frame and a confidence loss in the accuracy of the presence of the target, including wormholes, tears, surface mildew, internal corrosion; n _pos represents the number of positive samples of the appearance image and the internal image of the chestnut in the training data set; n _neg represents the number of negative samples of the appearance image and the internal image of the chestnut in the training data set; pi represents a confidence predictive value of an ith target detection frame in model output of the chestnut recognition model, and i represents a number of the target detection frame; gamma is an adjustable parameter for adjusting the degradation rate of FocalLoss, i.e. for controlling the weights of the easily classified samples and the difficult classified samples.

Further, the confidence loss L _conf for calculating whether a target exists within the target detection frame and the accuracy of the existence of the target can be expressed as follows:

Preferably, the weights of the easily classified samples and the difficult classified samples are adjusted using a weighted loss function.

Specifically, the loss value predicted by the chestnut recognition model is calculated by adopting the following formula:

PL(p_t)＝-(1-p_t)^γlog(p_t)

Wherein PL (p _t) represents a loss value predicted by the chestnut recognition model calculated by using a weighted loss function for adjusting weights of the easy-classification samples and the difficult-classification samples, and p _t represents a confidence score of the chestnut recognition model prediction; gamma denotes an adjustable parameter for controlling the weights of the easy-to-classify samples and the difficult-to-classify samples, takes a positive value, and when the value of gamma is zero, the Focal Loss is degenerated into a common cross entropy Loss function.

In a specific embodiment, for example, the appearance image of the chestnut and the internal image of the chestnut to be processed are input into a trained chestnut sorting and identifying model, and multidimensional vectors are output to indicate the size of the chestnut, whether the chestnut has worm eyes, whether the chestnut has surface mildew, the size of an internal mildew area and the like. The multidimensional vector= [ size, whether there is worm eye, whether there is surface mildew, internal mildew area size, color features, shape features, texture features, … ]. Wherein, the size refers to the size of the Chinese chestnut, and concretely refers to the dimensions of the Chinese chestnut such as the length, the width, the height and the like. The Chinese chestnut with or without the wormholes is specifically expressed by a binary value, 1 is expressed by the wormholes, and 0 is expressed by the absence of the wormholes. Whether the chestnut has surface mildew or not is indicated by a binary value. The 'size of the inner mildew area' refers to the size of the inner mildew area of the chestnut, and can be a numerical value which refers to the number of pixels or the relative proportion of the mildew area. "color characteristics" means color characteristics of chestnut, for example, a set of numerical values are used to represent color distribution characteristics of chestnut. "shape characteristics" means shape characteristics of chestnut, for example, a set of values are used to represent shape descriptive characteristics of chestnut, such as roundness, aspect ratio, etc. "texture features" means texture features of chestnut, for example, using a set of values to represent texture descriptive features of chestnut surface, such as smoothness, roughness, etc. The multidimensional vector can comprehensively represent various attributes of the Chinese chestnut, so that the Chinese chestnut can be more accurately sorted and identified.

In step S103, according to the identification and classification result, a sorting strategy is determined, and the chestnut to be processed is blown off the sorting conveying channel by using the blowing and separating device to fall to the corresponding blanking channel so as to complete the sorting operation.

And (5) carrying out chestnut recognition by using the optimized chestnut recognition model to obtain a recognition classification result. And then, the automatic sorting system of the Chinese chestnut formulates a sorting strategy based on the identification and sorting result so as to select a corresponding blanking groove for each Chinese chestnut to be processed. The sorting strategy specifically comprises the following contents: the interval from photographing to blowing is calculated, and the position information of each chestnut to be processed is combined to generate a blowing command, wherein the blowing command comprises a pulse value for transmission, the pulse value sorting conveyor belt transmits a pulse value to a chestnut automatic sorting system, for example, plc, and the chestnut automatic sorting system can calculate the position of each chestnut to be processed relative to the sorting conveyor belt according to the pulse value.

According to the blowing and separating command, the blowing and separating device is controlled to blow and separate the Chinese chestnut to be processed on the sorting conveyor belt in the sorting conveyor belt to the corresponding blanking channel, so that the sorting action of the Chinese chestnut to be processed is finished jointly through multiparty cooperation. Specifically when the chinese chestnut on the letter sorting conveyer belt conveys the corresponding position, the control module control of chinese chestnut automation letter sorting system's host computer blows off the letter sorting conveying passageway with waiting to handle chinese chestnut and blows off to drop to corresponding silo way in order to accomplish the letter sorting operation.

In an alternative embodiment, the discharge channels are used, each corresponding to a category of chestnut. The chestnut after being blown and separated is distributed to corresponding blanking channels according to the size and the quality

It should be noted that, the content of the automatic chestnut sorting system according to the first aspect of the present invention is substantially the same as that of the automatic chestnut sorting system according to the second aspect of the present invention, and therefore, the description of the same parts is omitted. Furthermore, the drawings are only schematic illustrations of processes involved in a method according to an exemplary embodiment of the present invention, and are not intended to be limiting. It will be readily understood that the processes shown in the figures do not indicate or limit the temporal order of these processes. In addition, it is also readily understood that these processes may be performed synchronously or asynchronously, for example, among a plurality of modules.

Compared with the prior art, the invention utilizes the dispersion feeding mechanism to convey the feeding conveyor belt loaded with the Chinese chestnut to be processed to the dispersion roller so as to disperse the Chinese chestnut to be processed to a plurality of sorting conveying channels; monitoring sorting and conveying conditions in real time, identifying position information of the chestnut to be processed on the sorting and conveying channels, and collecting appearance images and internal images of the chestnut; according to the collected appearance images and internal images of the Chinese chestnut, the appearance features and the internal features of the Chinese chestnut to be processed are identified by adopting an optimized Chinese chestnut identification model, so that an identification classification result can be accurately determined; according to the identification and classification result, the chestnut to be processed is blown off the sorting conveying channel by the blowing device so as to fall to the corresponding blanking channel to finish sorting operation, so that a more intelligent, more efficient and more automatic sorting process can be realized.

In addition, the invention improves the loss function of the chestnut recognition model, introduces FocalLoss, can make the chestnut recognition model pay more attention to the difficult-to-classify image samples, can improve the learning ability of the model to the difficult-to-classify samples, and further effectively solves the problem of unbalanced classification in target detection; the optimized chestnut recognition model can accurately describe the position, the size and the rotation angle of the target detection frame, and further can accurately detect and recognize the target.

In addition, the number of parameters and the calculated amount of the model are effectively reduced by adopting a MobileNetV depth separable convolution structure; in the task of chestnut identification, images can be processed more quickly, so that the identification process is accelerated; the method has the advantages that the calculated amount and the parameter quantity are reduced, and meanwhile, the higher recognition performance can be kept, namely, in the Chinese chestnut recognition task, chinese chestnuts in various different states can be effectively recognized, and multi-category classification can be accurately carried out; the depth separable convolution structure is suitable for the features with different scales and complexity, has good adaptability to the features with different scales and complexity, and particularly can effectively capture the defect features with different scales and complexity in the image to be processed in the chestnut recognition task, so that accurate classification and recognition are realized.

In addition, for the sorting and conveying process, visual interfaces which are visual and user-friendly are designed, so that operators can easily configure system parameters, monitor sorting conditions and perform manual intervention, and the flexibility and usability of the whole system are improved.

It should be noted that the foregoing detailed description is exemplary and is intended to provide further explanation of the application. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present application. As used herein, the singular is intended to include the plural unless the context clearly indicates otherwise. Furthermore, it will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, steps, operations, devices, components, and/or groups thereof.

It should be noted that the terms "first," "second," and the like in the description and the claims of the present application and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances such that the embodiments of the application described herein are capable of operation in sequences other than those illustrated or otherwise described herein.

Furthermore, the terms "comprise" and "have," as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those elements but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

In the above detailed description, reference is made to the accompanying drawings, which form a part hereof. In the drawings, like numerals typically identify like components unless context indicates otherwise. The illustrated embodiments described in the detailed description, drawings, and claims are not meant to be limiting. Other embodiments may be utilized, and other changes may be made, without departing from the spirit or scope of the subject matter presented herein.

The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. The utility model provides a chinese chestnut automatic sorting system which characterized in that includes:

The dispersing and feeding mechanism is used for loading and conveying the Chinese chestnut to be processed and comprises a hopper, a feeding conveyor belt, a dispersing roller, a redundant channel and a circulating conveyor belt, wherein the dispersing roller is a flexible self-rotating device, and the Chinese chestnut loaded on the dispersing roller is dispersed through self-rotation and is dispersed to the sorting conveyor channel;

The sorting mechanism is used for executing sorting action of the Chinese chestnut to be processed and comprises a sorting conveyor belt, a position identification device, a pneumatic blowing device and a blanking channel;

the identification mechanism is electrically connected with the sorting mechanism and is positioned at the front end of the sorting conveyor belt and used for collecting surface images and internal images of the Chinese chestnut to be processed on the sorting conveyor channel to identify the appearance characteristics and the internal characteristics of each Chinese chestnut to be processed, and the identification mechanism comprises a host, an industrial camera, X-ray imaging equipment, a light supplementing lamp and a lamp box, wherein the host is provided with an intelligent sorting module and a control module, the intelligent sorting module can automatically analyze the appearance characteristics and the internal characteristics of each Chinese chestnut to be processed, and the control module generates a blowing command according to sorting information of the intelligent sorting module so as to control the sorting action of the Chinese chestnut to be processed;

And the frame is provided with the dispersion feeding mechanism, the sorting mechanism and the identification mechanism.

2. The automated chestnut sorting system according to claim 1, wherein said system comprises:

The dispersion feeding mechanism is provided with at least four dispersion rollers, and each dispersion roller is provided with a flexible surface to contact the Chinese chestnut to be treated, so that the Chinese chestnut to be treated is uniformly dispersed to the sorting conveyor belt of the sorting mechanism through rotation.

3. The automatic chestnut sorting system according to claim 1, wherein,

And according to the blowing and separating command device, the sorting conveyor belt, the pneumatic blowing and separating device and the blanking channel are matched with each other to jointly finish the sorting action of the Chinese chestnut to be processed.

4. The automated chestnut sorting system according to claim 1, further comprising a friction bar positioned below the sorting conveyor,

The sorting conveyor belt comprises a chain and carrier rollers, wherein one chain corresponds to one group of gears, a pair of carrier rollers are symmetrically arranged on two sides of one chain along the conveying direction to form two sorting conveying channels, and each pair of carrier rollers is connected by a connecting shaft;

The friction strips are fixed on the lower side of the center of the bottom of the sorting conveyor belt, so that when the carrier rollers move along with the sorting conveyor belt, and each carrier roller starts to rotate under the action of friction force generated by each carrier roller and each friction strip when the carrier roller passes through the friction strips, the to-be-processed chestnut located on each carrier roller position on the sorting conveyor channel is driven to rotate along with the rotation of the corresponding carrier roller, and a picture of 360 rotations of the to-be-processed chestnut around the center thereof is taken and obtained.

5. The automatic chestnut sorting system according to claim 4, wherein,

Each idler has the following shape: the middle part is concave, and two side parts which extend outwards from the middle part to two ends are arranged on the middle part, so that a material carrying position is formed at a gap part between two adjacent carrier rollers in the conveying direction of the sorting conveyor belt and is used for carrying each chestnut to be processed.

6. The automatic chestnut sorting system according to claim 4, wherein,

The redundant channel is positioned below the sorting conveyor belt and is used for collecting scattered Chinese chestnut to be processed;

The first end of the circulating conveyor belt is connected with the redundant channel, the other end of the circulating conveyor belt is connected with the feeding conveyor belt, and the circulating conveyor belt is used for conveying the Chinese chestnut to be processed collected from the redundant channel back to the hopper.

7. The automatic chestnut sorting system according to claim 1, wherein,

A plurality of discharging channels are installed at the lower end of the frame, are uniformly distributed at equal intervals, and have both ends outside the both ends of the frame.

8. The automatic chestnut sorting system according to claim 1, wherein,

The machine frame comprises three part frames corresponding to the dispersing feeding mechanism, the sorting mechanism and the identification mechanism, and the three part frames are connected with each other and integrally formed; or the three partial frames are independent frames and are mutually connected.

9. A method for automatically sorting chinese chestnut, which performs automatic sorting of chinese chestnut using the automatic sorting system of any one of claims 1 to 8, comprising:

the Chinese chestnut to be processed in the hopper is transmitted to a dispersing roller of a dispersing and feeding mechanism, and the Chinese chestnut to be processed is dispersed by the self-rotation of the dispersing roller and is dispersed to a sorting and conveying channel;

Collecting surface images and internal images of the Chinese chestnut to be processed on the sorting and conveying channel, and identifying appearance features and internal features of the Chinese chestnut to be processed by adopting a pre-trained Chinese chestnut identification model to determine identification and classification results, wherein the method specifically comprises the following steps of: optimizing the target detection frame position loss, confidence loss, category loss and angle regression loss by improving a loss function so as to optimize the chestnut recognition model;

10. The automatic sorting method of chestnut according to claim 9, wherein,

Optimizing the chestnut recognition model by improving a loss function to optimize target detection frame position loss, confidence loss, category loss, angle regression loss, comprising:

L＝L_coord+L_conf+L_cls+λL_angle