CN115721296A - Wild animal identification system based on PaddleDeprotection - Google Patents

Abstract

The present invention relates to the technical field of long-term monitoring of wild animals, in particular to a wild animal identification system based on PaddleDetection. The system includes: several identification units for collecting image information of wild animals to be identified; a cloud platform for storing There are preset contour features matched with various types of wild animals and corresponding joint node information; a database is used to store preset contour features matched with various types of wild animals existing in the area where the identification unit is located; the central control A unit for preliminarily determining the type of wild animals existing in the area according to the area where each identification unit is located, and downloading the preset contour features existing in the area and corresponding joint node information to the database. The present invention effectively improves the recognition efficiency of the system of the present invention by judging whether the recognized contour features and the number of joint nodes conform to the distribution standard.

Description

A Wild Animal Recognition System Based on PaddleDetection

technical field

The invention relates to the technical field of long-term monitoring of wild animals, in particular to a PaddleDetection-based wild animal recognition system.

Background technique

Paddle is an open source deep learning platform independently developed by Baidu, which integrates the core framework of deep learning, tool components and service platforms. PaddleDetection is an excellent target detection development kit under the Paddle deep learning platform. It provides a variety of mainstream target detection, instance segmentation, and key point detection algorithms. It also modularizes each network component, provides data enhancement strategies, and loss function strategies. Model compression and cross-platform high-performance deployment can help industrial projects to be better implemented.

Long-term monitoring and identification of wildlife diversity is a key link in wildlife management, conservation, research and resource utilization. At present, the methods for monitoring wild animals mainly include line transect method, trap method and infrared camera shooting method.

Chinese patent publication number: CN114305389A discloses a wild animal detection and species identification system and method, the system includes: a first data acquisition unit, the first data acquisition unit includes a pressure detection component, and the pressure detection component is laid on On the ground, the first data acquisition unit outputs a trigger signal when the animal exerts pressure on the pressure detection component; the photographing unit, after receiving the trigger signal, the photographing unit takes pictures of the ground where the pressure detection component is located and outputs a photo ; a data processing unit, the data processing unit processes the photo and outputs at least a corresponding animal category. When a wild animal crawls, jumps, walks or runs on the pressure detection component, the pressure detection component will receive the pressure of the wild animal to generate a trigger signal.

The existing technology cannot determine the animal species to be recognized according to the contour features in the recognized image and the distribution of joint nodes, resulting in low recognition efficiency.

Contents of the invention

For this reason, the present invention provides a wild animal recognition system based on PaddleDetection to overcome the problem in the prior art that the recognition efficiency cannot be low due to the similarity of contours and the number of joint nodes meeting the standard.

To achieve the above object, the wild animal recognition system based on PaddleDetection of the present invention includes:

A plurality of identification units, which are respectively arranged at corresponding positions in different areas, are used to collect image information of wild animals to be identified;

The cloud platform is used to store preset contour features and corresponding joint node information matched with various wild animal species;

A database for storing preset contour features matched with various types of wild animals existing in the area where the identification unit is located, corresponding joint node information, and image information collected by the identification unit;

The central control unit is connected to each of the identification units, the database and the cloud platform, and is used to preliminarily determine the types of wild animals existing in the area according to the area where each of the identification units is located so that the cloud The preset contour features matched with various types of wild animals existing in the area in the platform and the corresponding joint node information are downloaded to the database, and the contour features obtained from the image information collected by the recognition unit are combined with Each preset outline feature in the database is matched to make a preliminary determination of whether the outline feature is a wild animal of the corresponding type, and when the outline feature is initially determined to be a wild animal of the corresponding type, according to the prediction of the wild animal of the type, Set the joint node information to further determine whether the contour feature is a wild animal of this type.

Further, the central control unit matches the contour features obtained from the images collected by the recognition unit with the preset contour features in the database, and makes statistics to obtain the preset contour features closest to the contour features. Outline feature and obtain the outline similarity S between the outline feature and the preset outline feature, and the central control unit compares S with the preset standard similarity S0 set by the central control unit to determine whether the outline feature is the corresponding type Preliminary identification of wild animals,

If S≥S0, the central control unit judges that the similarity between the contour feature and the preset contour feature closest to the contour feature meets the standard, and the central control unit initially judges the contour feature as a wild animal, and compares the wild animal The type of animal is preliminarily determined as the type of wild animal to which the preset contour feature closest to the contour feature belongs; the central control unit identifies and collects the joint node information in the contour feature and The preset joint node information of the contour feature is further judged whether it is a wild animal of this type;

If S<S0, the central control unit determines that the similarity between the profile feature and the preset profile feature closest to the profile feature does not meet the standard, and the central control unit compares the profile feature with each preset profile feature in the cloud platform. Set the contour feature to match to further determine whether the contour feature is the corresponding type of wild animal.

Further, the central control unit establishes a coordinate system under the first preset condition, and collects joint node information in the contour features in the image information to generate corresponding coordinate values for each joint node, and the central control unit converts each joint The coordinate values of the nodes are respectively matched with the preset joint node information corresponding to the type of animal in the database, and the preset joint node information with the largest number of joint points in the joint node information is counted and obtained. The number E of the joint nodes in the above joint node information that do not coincide with the preset joint nodes in the preset joint node information is further judged whether the contour feature is a wild animal of this type, and the central control unit is provided with a first preset that does not meet The number E1 of the distribution standard joint nodes, the second preset number E2 of the joint nodes that do not meet the distribution standard, and the third preset number E3 of the joint nodes that do not meet the distribution standard, wherein, E1<E2<E3,

If E<E1, the central control unit determines the outline feature as a corresponding type of wild animal;

If E1≤E<E2, the central control unit cannot determine whether the contour feature is a corresponding type of wild animal, the central control unit combines the contour features to obtain the priority of each joint node in the joint node information, and according to each The priority of the joint node further determines whether the contour feature is a specific type of wild animal;

If E2≤E<E3, the central control unit cannot determine whether the contour feature is a wild animal of the corresponding type, and the central control unit controls another recognition unit to collect image information of an object corresponding to the contour feature at another angle To obtain the corresponding second contour feature therefrom, and further determine whether the object corresponding to the contour feature is a wild animal of the corresponding type according to the second contour feature;

If E≥E3, the central control unit determines that the object corresponding to the contour feature is not a wild animal;

The first preset condition is that the central control unit determines that the contour similarity S between the contour feature and the preset contour feature satisfies S≥S0.

Further, the central control unit counts the number of non-overlapping joint nodes under the second preset condition so as to further determine whether the object corresponding to the joint node is a wild animal of the corresponding type according to the number R of non-overlapping joint nodes. Judgment, the central control unit has a preset number R0 of non-overlapping joint nodes,

If R≤R0, the central control unit determines that the object corresponding to the joint node is a wild animal of the corresponding type;

If R>R0, the central control unit determines that the object corresponding to the joint node is not a wild animal of the corresponding type;

The second preset condition is that the central control unit judges that the contour similarity S between the contour feature and the preset contour feature satisfies S≥S0 and the preset joints in the joint node information and the preset joint node information The number E of joint nodes whose nodes do not overlap satisfies E1≤E<E2.

Further, the central control unit counts the coordinates of the non-overlapping joint nodes and the level to which each joint node belongs, and determines whether to use the coordinates of the non-overlapping joint nodes according to the absolute distance H In the cloud platform, the preset contour features and preset joint node information corresponding to the new wild animal species are obtained to re-identify the contour features, and the central control unit is provided with a preset distance H0,

If H≤H0, the central control unit determines that it is not necessary to obtain new preset contour features and preset joint node information from the cloud platform;

If H>H0, the central control unit determines to obtain new preset contour features and preset joint node information from the cloud platform.

Further, for a single preset joint node, the central control unit detects the maximum movable angle θ of the connection between the joint node and the joint node as the motion center of the joint node to rate the joint node, and the central control unit according to the level To adjust the preset distance H0, the central control unit is provided with a first preset movable angle θ1, a second preset movable angle θ2 and a third preset movable angle θ3, wherein θ1<θ2<θ3,

If θ≤θ1, the central control unit determines that the joint node is a first-level joint node, and determines that there is no need to adjust the preset distance H0;

If θ1<θ≤θ2, the central control unit determines that the joint node is a secondary joint node, and uses e3 to adjust the preset distance H0 to a corresponding value;

If θ2<θ≤θ3, the central control unit determines that the joint node is a third-level joint node, and uses e2 to adjust the preset distance H0 to a corresponding value;

If θ>θ3, the central control unit determines that the joint node is a fourth-level joint node, and uses e1 to adjust the preset distance H0 to a corresponding value;

The central control unit records the preset distance adjusted by ej as H0', and sets H0'=H0×ej, where j=1, 2, 3.

Further, under the third preset condition, the central control unit counts the joint node information corresponding to the second contour feature collected by other recognition units in the area, and collects joint node information that is not related to the second contour feature. The identification units whose number E of the joint nodes in the joint node information coincident with the corresponding preset joint nodes are greater than the preset value are recorded as qualified identification units, and the central control unit compares the number of qualified identification units with the number of identification units whose contours are collected. The number ratio is recorded as D and compared with the preset standard ratio D0 set in the central control unit to determine whether the object corresponding to the outline feature is a wild animal of the corresponding type,

If D≥D0, the central control unit determines that the object corresponding to the second contour feature is a wild animal of a corresponding type;

If D<D0, the central control unit determines that the object corresponding to the second contour feature is not a wild animal of the corresponding type;

The third preset condition is that the central control unit judges that the contour similarity S between the contour feature and the preset contour feature satisfies S≥S0 and the preset joints in the joint node information and the preset joint node information The number E of joint nodes whose nodes do not overlap satisfies E2≤E<E3.

Further, under the fourth preset condition, the central control unit retrieves from the cloud platform based on the contour features collected by each of the recognition units and the corresponding joint node information, so as to determine whether the object is for a particular species;

The fourth preset condition is that the central control unit judges that the contour similarity S between the contour feature and the preset contour feature satisfies S≥S0, and the preset joints in the joint node information and the preset joint node information The number E of non-overlapping joint nodes satisfies E2≤E<E3, and the absolute distance H between the coordinates of non-overlapping joint nodes and the coordinates of the corresponding preset joint nodes satisfies H>H0 and the number of qualified recognition units is the same as the collected The proportion D of the number of recognition units of the contour satisfies D<D0.

Further, the central control unit analyzes the contour feature information collected by each of the recognition units within a preset period according to the recognition result, and transmits the contour feature information to the cloud platform to update the cloud platform for the type of animal contour The identification standard of the feature, the central control unit calculates the ratio B of the number of completed identification and the number of captured images, and compares B with each preset ratio, and adjusts the standard similarity to the corresponding value according to the comparison result, the central control unit There are first preset ratio B1, second preset ratio B2, first preset standard similarity adjustment coefficient α1, second preset standard similarity adjustment coefficient α2 and third preset standard similarity adjustment coefficient α3, wherein , B1<B2, 1<α1<α2<α3<1.4,

If B≤B1, the central control unit uses α1 to adjust the standard similarity to a corresponding value, and the adjusted standard similarity is denoted as S', and S'=S×α3 is set;

If B1<B≤B2, the central control unit uses α2 to adjust the standard similarity to a corresponding value, and the adjusted standard similarity is denoted as S', and S'=S×α2 is set;

If B>B2, the central control unit uses α3 to adjust the standard similarity to a corresponding value, and the adjusted standard similarity is denoted as S', and S'=S×α1 is set.

Further, the central control unit determines the usage amount of the recognition unit according to the vegetation density G of the region to be detected, and the central control unit is provided with a first preset density G1, a second preset density G2, a first preset usage amount The adjustment coefficient β1, the second preset usage adjustment coefficient β2 and the third preset usage adjustment coefficient β3, wherein G1<G2, 1<β1<β2<β3<1.6,

If G≤G1, the central control unit uses β1 to adjust the usage amount of the recognition unit to a corresponding value, and the adjusted usage amount of the recognition unit is denoted as W', set W'=W×(β1-1) , where W is the initial usage of the recognition unit;

If G1<G≤G2, the central control unit uses β2 to adjust the usage amount of the recognition unit to a corresponding value, and the adjusted usage amount of the recognition unit is recorded as W', and W'=W×β2 is set;

If G>G2, the central control unit uses β3 to adjust the usage amount of the identification unit to a corresponding value, and the adjusted usage amount of the identification unit is denoted as W', and W'=W×β3 is set.

Compared with the prior art, the beneficial effect of the present invention is that the present invention compares the similarity of the obtained outline features with the preset standard similarity, and judges whether the outline features are wild animals of the corresponding type according to the comparison results. Judgment effectively improves the recognition efficiency of the system described in the present invention; at the same time, the present invention further determines the type of animal according to the distribution of joint nodes and the number of joint nodes that meet the distribution standard, and effectively further confirms the attributes of the animal to be recognized , to avoid the occurrence of recognition errors only due to similar contours, and further improve the recognition efficiency of the system of the present invention.

Further, the central control unit of the present invention is provided with a preset standard similarity S0, by comparing the collected contour features with the preset contour features, the contour similarity is determined, and the contour similarity is compared with the preset standard similarity By making a comparison, it is judged whether the contour features meet the standard, and the recognition efficiency of the system of the present invention is effectively improved by preliminarily judging the animal species from the contour.

Further, the central control unit of the present invention is provided with a number of preset joint nodes that do not meet the distribution standard, by comparing the number of joint nodes that do not meet the distribution standard with the number of each preset joint node that does not meet the distribution standard In contrast, it is determined whether the contour feature is a wild animal of this type, thereby further determining the type of animal to be identified, and further improving the identification efficiency of the system of the present invention.

Further, the central control unit of the present invention is provided with a preset standard proportion D0, and the identification unit whose number E of the joint nodes in the joint node information of which the joint nodes are not overlapped is greater than the preset value is recorded as a qualified identification unit , record the ratio of the number of qualified recognition units to the number of recognition units that have collected contours as D, and compare D with D0 to further determine whether the object corresponding to the contour feature is a wild animal of the corresponding type according to the comparison result. By judging the object corresponding to the outline feature from multiple angles, the recognition efficiency of the system of the present invention is effectively improved, and the misjudgment caused by only referring to the image of a single angle is avoided, thereby affecting the recognition efficiency.

Further, the present invention judges that the contour similarity is lower than the preset standard similarity, the proportion of the recognition units whose joint nodes meet the distribution standard is greater than the preset value is lower than the preset standard proportion, and the non-overlapping joint nodes When the absolute distance between the coordinates and the coordinates of the corresponding preset joint nodes is greater than the preset distance and the ratio of the number of qualified recognition units to the number of recognition units collected from the contour is less than the preset ratio, the collected contour features Retrieve with each joint node through the cloud platform to further determine the type of animal to be identified, and further determine the type of animal to be identified with the help of network big data, effectively improving the identification efficiency of the system of the present invention.

Further, the central control unit of the present invention is provided with a number of preset ratios and a number of preset standard similarity adjustment coefficients. By comparing the ratio B of the number of recognized images to the number of captured images with the preset ratio, according to the comparison result By adjusting the preset standard similarity to a corresponding value, the recognition accuracy is effectively improved by adjusting the preset standard similarity, and the recognition efficiency of the system of the present invention is further improved.

Description of drawings

Fig. 1 is the structural block diagram of the wild animal recognition system based on PaddleDetection of the embodiment of the present invention;

Fig. 2 is a flow chart for further determining whether the contour feature is a wild animal of this type according to the number E of non-overlapping joint nodes according to the embodiment of the present invention;

Fig. 3 is a flow chart of determining whether the object corresponding to the contour feature is a wild animal of the corresponding type according to the ratio of the number of qualified recognition units to the number of recognition units that have collected the contour;

Fig. 4 is a flow chart of adjusting the similarity of preset standards according to the ratio of the number of recognized images to the number of captured images according to an embodiment of the present invention.

Detailed ways

In order to make the objects and advantages of the present invention clearer, the present invention will be further described below in conjunction with the examples; it should be understood that the specific examples described here are only for explaining the present invention, and are not intended to limit the present invention.

Preferred embodiments of the present invention are described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principle of the present invention, and are not intended to limit the protection scope of the present invention.

It should be noted that, in the description of the present invention, terms such as "upper", "lower", "left", "right", "inner", "outer" and other indicated directions or positional relationships are based on the terms shown in the accompanying drawings. The direction or positional relationship shown is only for convenience of description, and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention.

In addition, it should be noted that, in the description of the present invention, unless otherwise clearly stipulated and limited, the terms "installation", "connection" and "connection" should be understood in a broad sense, for example, it can be a fixed connection or a It is a detachable connection or an integral connection; it may be a mechanical connection or an electrical connection; it may be a direct connection or an indirect connection through an intermediary, and it may be the internal communication of two components. Those skilled in the art can understand the specific meanings of the above terms in the present invention according to specific situations.

See also shown in Fig. 1, it is the structural block diagram of the wild animal recognition system based on PaddleDetection of the embodiment of the present invention, the wild animal recognition system based on PaddleDetection of the present invention, comprises:

A plurality of identification units, which are respectively arranged at corresponding positions in different areas, are used to collect image information of wild animals to be identified;

The cloud platform is used to store preset contour features and corresponding joint node information matched with various wild animal species;

A database for storing preset contour features matched with various types of wild animals existing in the area where the identification unit is located, corresponding joint node information, and image information collected by the identification unit;

The central control unit is connected to each of the identification units, the database and the cloud platform, and is used to preliminarily determine the types of wild animals existing in the area according to the area where each of the identification units is located so that the cloud The preset contour features matched with various types of wild animals existing in the area in the platform and the corresponding joint node information are downloaded to the database, and the contour features obtained from the image information collected by the recognition unit are combined with Each preset outline feature in the database is matched to make a preliminary determination of whether the outline feature is a wild animal of the corresponding type, and when the outline feature is initially determined to be a wild animal of the corresponding type, according to the prediction of the wild animal of the type, Set the joint node information to further determine whether the contour feature is a wild animal of this type.

Specifically, the central control unit matches the contour features obtained from the images collected by the recognition unit with the preset contour features in the database, and calculates and obtains the preset contour features that are most similar to the contour features. Set the contour feature and obtain the contour similarity S between the contour feature and the preset contour feature, and the central control unit compares S with the preset standard similarity S0 set by the central control unit to determine whether the contour feature is corresponding Preliminary determination of species of wild animals,

If S≥S0, the central control unit judges that the similarity between the contour feature and the preset contour feature closest to the contour feature meets the standard, and the central control unit initially judges the contour feature as a wild animal, and compares the wild animal The type of animal is preliminarily determined as the type of wild animal to which the preset contour feature closest to the contour feature belongs; the central control unit identifies and collects the joint node information in the contour feature and The preset joint node information of the contour feature is further judged whether it is a wild animal of this type;

If S<S0, the central control unit determines that the similarity between the profile feature and the preset profile feature closest to the profile feature does not meet the standard, and the central control unit compares the profile feature with each preset profile feature in the cloud platform. Set the contour feature to match to further determine whether the contour feature is the corresponding type of wild animal.

It can be understood that the calculation method for the similarity between the contour features obtained from the image information and the corresponding preset standard contour features is: counting the length of the completely overlapping contour lines in the two contours and the total length of the corresponding preset standard contour Proportion.

The central control unit of the present invention is provided with a preset standard similarity S0, by comparing the collected contour features with the preset contour features, the contour similarity is determined, and the contour similarity is compared with the preset standard similarity, To judge whether the contour features meet the standard, the recognition efficiency of the system of the present invention is effectively improved by preliminarily judging the animal species from the contour.

Please refer to Fig. 2, which is a flow chart of the embodiment of the present invention to determine the animal species to be identified according to the number of joint nodes that do not meet the distribution standard. The central control unit establishes a coordinate system under the first preset condition, and collects The joint node information in the contour feature in the image information is used to generate corresponding coordinate values for each joint node, and the central control unit compares the coordinate values of each joint node with each preset joint node corresponding to the type of animal in the database Information matching, counting and obtaining the preset joint node information with the largest number of overlapping joint points in the joint node information, and according to the joint node information in the joint node information that does not overlap with the preset joint node information in the preset joint node information The number E of the nodes further determines whether the contour feature is a wild animal of this type, and the central control unit sets the first preset number E1 of joint nodes that do not meet the distribution standard, the second preset number of joint nodes that do not meet the distribution standard The quantity E2 and the third preset quantity E3 of joint nodes that do not meet the distribution standard, wherein, E1<E2<E3,

If E<E1, the central control unit determines the outline feature as a corresponding type of wild animal;

If E1≤E<E2, the central control unit cannot determine whether the contour feature is a corresponding type of wild animal, the central control unit combines the contour features to obtain the priority of each joint node in the joint node information, and according to each The priority of the joint node further determines whether the contour feature is a specific type of wild animal;

If E2≤E<E3, the central control unit cannot determine whether the contour feature is a wild animal of the corresponding type, and the central control unit controls another recognition unit to collect image information of an object corresponding to the contour feature at another angle To obtain the corresponding second contour feature therefrom, and further determine whether the object corresponding to the contour feature is a wild animal of the corresponding type according to the second contour feature;

If E≥E3, the central control unit determines that the object corresponding to the contour feature is not a wild animal;

The first preset condition is that the central control unit determines that the contour similarity S between the contour feature and the preset contour feature satisfies S≥S0.

It can be understood that, regarding the preset joint node information, the program sequentially simulates all the movements of the animal to form multiple continuous joint node distribution points, and projects the joint nodes into the coordinate system to obtain a series of coordinate values.

The central control unit of the present invention is provided with a number of preset joint nodes that do not meet the distribution standard, and by comparing the number of joint nodes that do not meet the distribution standard with the number of each preset joint node that does not meet the distribution standard, it is determined Whether the outline feature is a wild animal of this type can further determine the type of animal to be identified, and further improve the identification efficiency of the system of the present invention.

Specifically, the central control unit counts the number of non-overlapping joint nodes under the second preset condition to determine whether the object corresponding to the joint node is a wild animal of the corresponding type according to the number R of non-overlapping joint nodes. It is further determined that the central control unit has a preset number of non-overlapping joint nodes R0,

If R≤R0, the central control unit determines that the object corresponding to the joint node is a wild animal of the corresponding type;

If R>R0, the central control unit determines that the object corresponding to the joint node is not a wild animal of the corresponding type;

The second preset condition is that the central control unit judges that the contour similarity S between the contour feature and the preset contour feature satisfies S≥S0 and the preset joints in the joint node information and the preset joint node information The number E of joint nodes whose nodes do not overlap satisfies E1≤E<E2.

Specifically, the central control unit counts the coordinates of the non-overlapping joint nodes and the level to which each joint node belongs, and determines whether to Obtain preset contour features and preset joint node information corresponding to new wild animal species from the cloud platform to re-identify the contour features, and the central control unit is provided with a preset distance H0,

If H≤H0, the central control unit determines that it is not necessary to obtain new preset contour features and preset joint node information from the cloud platform;

If H>H0, the central control unit determines to obtain new preset contour features and preset joint node information from the cloud platform.

Specifically, for a single preset joint node, the central control unit detects the maximum movable angle θ of the connection between the joint node and the joint node as the motion center of the joint node to rate the joint node, and the central control unit according to The level adjusts the preset distance H0, and the central control unit is provided with a first preset movable angle θ1, a second preset movable angle θ2 and a third preset movable angle θ3, wherein θ1<θ2<θ3,

If θ≤θ1, the central control unit determines that the joint node is a first-level joint node, and determines that there is no need to adjust the preset distance H0;

If θ1<θ≤θ2, the central control unit determines that the joint node is a secondary joint node, and uses e3 to adjust the preset distance H0 to a corresponding value;

If θ2<θ≤θ3, the central control unit determines that the joint node is a third-level joint node, and uses e2 to adjust the preset distance H0 to a corresponding value;

If θ>θ3, the central control unit determines that the joint node is a fourth-level joint node, and uses e1 to adjust the preset distance H0 to a corresponding value;

The central control unit records the preset distance adjusted by ej as H0', and sets H0'=H0×ej, where j=1, 2, 3.

Please refer to FIG. 3 , which is a flow chart showing whether the ratio of the number of qualified recognition units to the number of recognition units whose contours are collected meets the standard in the embodiment of the present invention. Under the third preset condition, the central control unit, Count the joint node information corresponding to the second contour feature collected by other recognition units in the area, and the number E of joint nodes that do not overlap with the corresponding preset joint nodes in the collected joint node information is greater than the preset The recognition unit with the highest value is recorded as a qualified recognition unit, and the central control unit records the ratio of the number of qualified recognition units and the number of recognition units that have collected contours as D, and performs the calculation according to D and the preset standard ratio D0 set in the central control unit. Compare to determine whether the object corresponding to the outline feature is a wild animal of the corresponding type,

If D≥D0, the central control unit determines that the object corresponding to the second contour feature is a wild animal of a corresponding type;

If D<D0, the central control unit determines that the object corresponding to the second contour feature is not a wild animal of the corresponding type;

The third preset condition is that the central control unit judges that the contour similarity S between the contour feature and the preset contour feature satisfies S≥S0 and the preset joints in the joint node information and the preset joint node information The number E of joint nodes whose nodes do not overlap satisfies E2≤E<E3.

The central control unit of the present invention is provided with a preset standard ratio D0, and the identification unit whose number E of the joint nodes in the joint node information that the joint nodes do not overlap is greater than the preset value is recorded as a qualified recognition unit, and the qualified The ratio of the number of recognition units to the number of recognition units collected from the outline is recorded as D. By comparing D with D0, it is further determined whether the object corresponding to the outline feature is a wild animal of the corresponding type according to the comparison result. Through multi-angle The determination of the object corresponding to the contour feature effectively improves the recognition efficiency of the system of the present invention, and avoids misjudgment caused by only referring to an image from a single angle, thereby affecting the recognition efficiency.

Specifically, under the fourth preset condition, the central control unit retrieves from the cloud platform based on the contour features collected by each of the recognition units and the corresponding joint node information to determine whether the object whether it is a specific species;

The fourth preset condition is that the central control unit judges that the contour similarity S between the contour feature and the preset contour feature satisfies S≥S0, and the preset joints in the joint node information and the preset joint node information The number E of non-overlapping joint nodes satisfies E2≤E<E3, and the absolute distance H between the coordinates of non-overlapping joint nodes and the coordinates of the corresponding preset joint nodes satisfies H>H0 and the number of qualified recognition units is the same as the collected The proportion D of the number of recognition units of the contour satisfies D<D0.

The present invention judges that the contour similarity is lower than the preset standard similarity, the proportion of the recognition units whose joint nodes meet the distribution standard is greater than the preset value is lower than the preset standard proportion, and the coordinates of the non-overlapping joint nodes and the corresponding When the absolute distance of the coordinates of the preset joint nodes is greater than the preset distance and the ratio of the number of qualified recognition units to the number of collected contour recognition units is less than the preset ratio, by combining the collected contour features with each joint The nodes search through the cloud platform to further determine the type of animal to be identified, and further determine the type of animal to be identified with the help of network big data, effectively improving the identification efficiency of the system of the present invention.

Please refer to FIG. 4 , which is a flow chart of adjusting the similarity of preset standards according to the ratio of the number of completed recognitions to the number of captured images according to an embodiment of the present invention. The central control unit analyzes each of the recognition units according to the recognition results The profile feature information collected within the preset period, and the profile feature information is sent to the cloud platform to update the cloud platform's identification standard for the profile feature of this type of animal, and the central control unit calculates the number of completed identification and the number of collected images ratio B, and compare B with each preset ratio, adjust the standard similarity to the corresponding value according to the comparison result, the central control unit is provided with a first preset ratio B1, a second preset ratio B2, a first The preset standard similarity adjustment coefficient α1, the second preset standard similarity adjustment coefficient α2 and the third preset standard similarity adjustment coefficient α3, wherein, B1<B2, 1<α1<α2<α3<1.4,

If B≤B1, the central control unit uses α1 to adjust the standard similarity to a corresponding value, and the adjusted standard similarity is denoted as S', and S'=S×α3 is set;

If B1<B≤B2, the central control unit uses α2 to adjust the standard similarity to a corresponding value, and the adjusted standard similarity is denoted as S', and S'=S×α2 is set;

If B>B2, the central control unit uses α3 to adjust the standard similarity to a corresponding value, and the adjusted standard similarity is denoted as S', and S'=S×α1 is set.

The central control unit of the present invention is provided with a number of preset ratios and a number of preset standard similarity adjustment coefficients. By comparing the ratio B of the number of recognized images to the number of captured images with the preset ratio, the preset The standard similarity is adjusted to a corresponding value, and the recognition accuracy is effectively improved by adjusting the preset standard similarity, and the recognition efficiency of the system of the present invention is further improved.

Specifically, the central control unit determines the usage amount of the recognition unit according to the vegetation density G of the area to be detected, and the central control unit is provided with a first preset density G1, a second preset density G2, a first preset usage Volume adjustment coefficient β1, second preset usage volume adjustment coefficient β2 and third preset usage volume adjustment coefficient β3, wherein G1<G2, 1<β1<β2<β3<1.6,

If G≤G1, the central control unit uses β1 to adjust the usage amount of the recognition unit to a corresponding value, and the adjusted usage amount of the recognition unit is denoted as W', set W'=W×(β1-1) , where W is the initial usage of the recognition unit;

If G1<G≤G2, the central control unit uses β2 to adjust the usage amount of the recognition unit to a corresponding value, and the adjusted usage amount of the recognition unit is recorded as W', and W'=W×β2 is set;

If G>G2, the central control unit uses β3 to adjust the usage amount of the identification unit to a corresponding value, and the adjusted usage amount of the identification unit is denoted as W', and W'=W×β3 is set.

So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings, however, those skilled in the art will easily understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to related technical features, and the technical solutions after these changes or substitutions will all fall within the protection scope of the present invention.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention; for those skilled in the art, the present invention may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims (10)

1. A wild animal recognition system based on PaddleDetection, characterized in that, comprising: A plurality of identification units, which are respectively arranged at corresponding positions in different areas, are used to collect image information of wild animals to be identified; The cloud platform is used to store preset contour features and corresponding joint node information matched with various wild animal species; A database for storing preset contour features matched with various types of wild animals existing in the area where the identification unit is located, corresponding joint node information, and image information collected by the identification unit; The central control unit is connected to each of the identification units, the database and the cloud platform, and is used to preliminarily determine the types of wild animals existing in the area according to the area where each of the identification units is located so that the cloud The preset contour features matched with various types of wild animals existing in the area in the platform and the corresponding joint node information are downloaded to the database, and the contour features obtained from the image information collected by the recognition unit are combined with Each preset outline feature in the database is matched to make a preliminary determination of whether the outline feature is a wild animal of the corresponding type, and when the outline feature is initially determined to be a wild animal of the corresponding type, according to the prediction of the wild animal of the type, Set the joint node information to further determine whether the contour feature is a wild animal of this type. 2. The wild animal recognition system based on PaddleDetection according to claim 1, wherein the central control unit combines the contour features obtained from the image collected by the recognition unit with each preset in the database. The contour features are matched, the statistics are obtained and the preset contour features closest to the contour features are obtained, and the contour similarity S between the contour features and the preset contour features is obtained, and the central control unit compares S with the preset contour features set by the central control unit. The preset standard similarity S0 is compared to make a preliminary judgment on whether the contour feature is a wild animal of the corresponding type, If S≥S0, the central control unit judges that the similarity between the contour feature and the preset contour feature closest to the contour feature meets the standard, and the central control unit initially judges the contour feature as a wild animal, and compares the wild animal The type of animal is preliminarily determined as the type of wild animal to which the preset contour feature closest to the contour feature belongs; the central control unit identifies and collects the joint node information in the contour feature and The preset joint node information of the contour feature is further judged whether it is a wild animal of this type; If S<S0, the central control unit determines that the similarity between the profile feature and the preset profile feature closest to the profile feature does not meet the standard, and the central control unit compares the profile feature with each preset profile feature in the cloud platform. Set the contour feature to match to further determine whether the contour feature is the corresponding type of wild animal. 3. The wild animal recognition system based on PaddleDetection according to claim 2, wherein the central control unit establishes a coordinate system under the first preset condition, and collects the joint nodes in the contour features in the image information information to generate corresponding coordinate values for each joint node, and the central control unit matches the coordinate values of each joint node with the information of each preset joint node corresponding to this type of animal in the database, and calculates and obtains the corresponding coordinate value of the joint node. In the node information, the preset joint node information with the largest number of overlapping joint points, and according to the number E of joint nodes in the joint node information that do not overlap with the preset joint nodes in the preset joint node information, determine whether the contour feature is the The species of wild animals is further judged, and the central control unit sets the first preset number E1 of joint nodes that do not meet the distribution standard, the second preset number E2 of joint nodes that do not meet the distribution standard, and the third preset number of joint nodes that do not meet the distribution standard The number of joint nodes E3, where, E1<E2<E3, If E<E1, the central control unit determines the outline feature as a corresponding type of wild animal; If E1≤E<E2, the central control unit cannot determine whether the contour feature is a corresponding type of wild animal, the central control unit combines the contour features to obtain the priority of each joint node in the joint node information, and according to each The priority of the joint node further determines whether the contour feature is a specific type of wild animal; If E2≤E<E3, the central control unit cannot determine whether the contour feature is a wild animal of the corresponding type, and the central control unit controls another recognition unit to collect image information of an object corresponding to the contour feature at another angle To obtain the corresponding second contour feature therefrom, and further determine whether the object corresponding to the contour feature is a wild animal of the corresponding type according to the second contour feature; If E≥E3, the central control unit determines that the object corresponding to the contour feature is not a wild animal; The first preset condition is that the central control unit determines that the contour similarity S between the contour feature and the preset contour feature satisfies S≥S0. 4. the wild animal recognition system based on PaddleDetection according to claim 3, is characterized in that, described central control unit counts the quantity of the joint node that does not overlap under the second preset condition so that according to the quantity of the joint node that does not overlap R further judges whether the object corresponding to the joint node is a wild animal of the corresponding type, and the central control unit has a preset number of non-overlapping joint nodes R0, If R≤R0, the central control unit determines that the object corresponding to the joint node is a wild animal of the corresponding type; If R>R0, the central control unit determines that the object corresponding to the joint node is not a wild animal of the corresponding type; The second preset condition is that the central control unit judges that the contour similarity S between the contour feature and the preset contour feature satisfies S≥S0 and the preset joints in the joint node information and the preset joint node information The number E of joint nodes whose nodes do not overlap satisfies E1≤E<E2. 5. The wild animal recognition system based on PaddleDetection according to claim 4, wherein the central control unit counts the coordinates of each joint node that does not overlap and the level to which each joint node belongs, and according to the joint nodes that do not overlap The absolute distance H between the coordinates of the corresponding preset joint node and the coordinates of the corresponding preset joint node determines whether to obtain the preset contour feature corresponding to the new wild animal species and the preset joint node information from the cloud platform to reconfigure the contour feature identification, the central control unit has a preset distance H0, If H≤H0, the central control unit determines that it is not necessary to obtain new preset contour features and preset joint node information from the cloud platform; If H>H0, the central control unit determines to obtain new preset contour features and preset joint node information from the cloud platform. 6. The wild animal recognition system based on PaddleDetection according to claim 5, characterized in that, for a single preset joint node, the central control unit detects the connection between the joint node and the joint node as the center of motion of the joint node The maximum activity angle θ of the joint node is rated, and the central control unit adjusts the preset distance H0 according to the level. The central control unit has a first preset activity angle θ1, a second preset activity angle θ2, a second preset activity angle Three preset activity angles θ3, first preset distance adjustment coefficient e1, second preset distance adjustment coefficient e2 and third preset distance adjustment coefficient e3, where θ1<θ2<θ3, 0<e1<e2<e3<1 , If θ≤θ1, the central control unit determines that the joint node is a first-level joint node, and determines that there is no need to adjust the preset distance H0; If θ1<θ≤θ2, the central control unit determines that the joint node is a secondary joint node, and uses e3 to adjust the preset distance H0 to a corresponding value; If θ2<θ≤θ3, the central control unit determines that the joint node is a third-level joint node, and uses e2 to adjust the preset distance H0 to a corresponding value; If θ>θ3, the central control unit determines that the joint node is a fourth-level joint node, and uses e1 to adjust the preset distance H0 to a corresponding value; The central control unit records the preset distance adjusted by ej as H0', and sets H0'=H0×ej, where j=1, 2, 3. 7. The wild animal recognition system based on PaddleDetection according to claim 6, characterized in that, the central control unit, under the third preset condition, counts the data collected by other said recognition units in the area and said first The joint node information corresponding to the two contour features, the number E of joint nodes that do not coincide with the corresponding preset joint nodes in the collected joint node information is recorded as a qualified recognition unit, and the central control unit recognizes the qualified The ratio of the number of units to the number of recognition units that have collected contours is denoted as D, and is compared with the preset standard ratio D0 set in the central control unit according to D to determine whether the object corresponding to the contour feature is of the corresponding type wild animals, If D≥D0, the central control unit determines that the object corresponding to the second contour feature is a wild animal of a corresponding type; If D<D0, the central control unit determines that the object corresponding to the second contour feature is not a wild animal of the corresponding type; The third preset condition is that the central control unit judges that the contour similarity S between the contour feature and the preset contour feature satisfies S≥S0 and the preset joints in the joint node information and the preset joint node information The number E of joint nodes whose nodes do not overlap satisfies E2≤E<E3. 8. The wild animal recognition system based on PaddleDetection according to claim 7, characterized in that, the central control unit collects each profile feature and each corresponding joint according to each of the recognition units under the fourth preset condition Node information is used as a reference to retrieve from the cloud platform to determine whether the object is a specific species; The fourth preset condition is that the central control unit judges that the contour similarity S between the contour feature and the preset contour feature satisfies S≥S0, and the preset joints in the joint node information and the preset joint node information The number E of non-overlapping joint nodes satisfies E2≤E<E3, and the absolute distance H between the coordinates of non-overlapping joint nodes and the coordinates of the corresponding preset joint nodes satisfies H>H0 and the number of qualified recognition units is the same as the collected The proportion D of the number of recognition units of the contour satisfies D<D0. 9. The wild animal recognition system based on PaddleDetection according to claim 8, wherein the central control unit analyzes the profile feature information collected by each of the recognition units in a preset period according to the recognition result, and The profile feature information is sent to the cloud platform to update the cloud platform’s identification standard for the profile features of the species, and the central control unit calculates the ratio B of the number of completed identifications and the number of images collected, and compares B with each preset ratio , adjust the standard similarity to the corresponding value according to the comparison result, the central control unit is provided with a first preset ratio B1, a second preset ratio B2, a first preset standard similarity adjustment coefficient α1, a second preset standard Similarity adjustment coefficient α2 and third preset standard similarity adjustment coefficient α3, wherein, B1<B2, 1<α1<α2<α3<1.4, If B≤B1, the central control unit uses α1 to adjust the standard similarity to a corresponding value, and the adjusted standard similarity is denoted as S', and S'=S×α3 is set; If B1<B≤B2, the central control unit uses α2 to adjust the standard similarity to a corresponding value, and the adjusted standard similarity is denoted as S', and S'=S×α2 is set; If B>B2, the central control unit uses α3 to adjust the standard similarity to a corresponding value, and the adjusted standard similarity is denoted as S', and S'=S×α1 is set. 10. the wild animal identification system based on PaddleDetection according to claim 9, is characterized in that, described central control unit determines the usage quantity of described identification unit according to the vegetation density G of region to be detected, and central control unit is provided with first The preset density G1, the second preset density G2, the first preset usage adjustment coefficient β1, the second preset usage adjustment coefficient β2 and the third preset usage adjustment coefficient β3, where G1<G2, 1<β1 <β2<β3<1.6, If G≤G1, the central control unit uses β1 to adjust the usage amount of the recognition unit to a corresponding value, and the adjusted usage amount of the recognition unit is denoted as W', set W'=W×(β1-1) , where W is the initial usage of the recognition unit; If G1<G≤G2, the central control unit uses β2 to adjust the usage amount of the recognition unit to a corresponding value, and the adjusted usage amount of the recognition unit is recorded as W', and W'=W×β2 is set; If G>G2, the central control unit uses β3 to adjust the usage amount of the identification unit to a corresponding value, and the adjusted usage amount of the identification unit is denoted as W', and W'=W×β3 is set.

Images