Cordyceps sinensis field target identification device and method

Abstract

The present invention relates to the development and utilization of intelligent agricultural equipment technology. Disclosed are a device and method for identifying Cordyceps sinensis in the field. The device comprises an imaging device for capturing surface video images and a core control and computing module for controlling various components and analyzing video data. A transmission line connects the imaging device and the core control and computing module. The core control and computing module is connected to an alarm module for warning of the presence of Cordyceps sinensis and a power supply module for providing power to the device. The present invention facilitates the identification of Cordyceps sinensis in the field. Using the identification device to search for targets in the field improves the efficiency of field searches for Cordyceps sinensis, making them more accessible.

Description

Technical Field

[0001] The present invention relates to the field of development and utilization of intelligent agricultural equipment technology, and specifically to a cordyceps sinensis field target recognition device and method. Background Art

[0002] Cordyceps sinensis is a precious medicinal herb unique to China, known for its immune system modulating, anti-tumor, and fatigue-fighting properties. It is a fungus unique to the Qinghai-Tibet Plateau and has a restricted distribution, concentrated in areas above 3,000 meters above sea level. In my country, its distribution extends from the Qilian Mountains in the north to the high mountains of northwestern Yunnan in the south, from the mountainous western Sichuan plateau in the east to most of the Himalayas in the west, encompassing provinces and regions such as Tibet, Qinghai, Sichuan, Yunnan, and Gansu.

[0003] However, current Cordyceps harvesting is primarily done manually. From April to June each year, large numbers of migrant workers and laborers travel to the growing areas to dig. However, due to the sparse growth of Cordyceps sinensis, the small, above-ground grass, and the complex surface weeds, the digging process is like searching for a needle in a haystack. During the search, diggers must kneel or crawl on the damp mountainside, slowly searching with their eyesight before digging. This traditional method suffers from three serious problems: First, this purely manual method is slow and inefficient. Typically, a skilled digger can only uncover dozens of Cordyceps sinensis per day, requiring significant manpower. Second, due to the dampness of the Cordyceps meadows and the strong ultraviolet rays of the plateau, the prolonged crawling and searching process can be very damaging to the diggers' health. Third, the large-scale presence and activity on the meadows can significantly impact the grass and the environment, damaging the plateau's natural ecology. Summary of the Invention

[0004] To address the above issues, the present invention provides a device and method for identifying Cordyceps sinensis in the wild. The present invention facilitates the identification of Cordyceps sinensis in the wild, using the identification device to search for targets in the wild, thereby improving the efficiency of searching for Cordyceps sinensis in the wild and making it more convenient to search for Cordyceps sinensis.

[0005] In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0006] A cordyceps sinensis field target recognition device includes an imaging device for collecting surface scene images and a core control and computing module for controlling various components and analyzing scene images. The imaging device and the core control and computing module are electrically connected via a transmission line. The core control and computing module is connected to an alarm module for warning of the presence of cordyceps sinensis and a power supply module for providing power to the device.

[0007] As a preferred embodiment, the core control and computing module includes a circuit mainboard, a computing acceleration module, a computing chip and a storage device. The computing chip includes a first-level target detection module for outputting one or more candidate target areas where Cordyceps sinensis exists and a second-level regional target re-determination module for judging whether Cordyceps sinensis exists in the scene image. The first-level target detection module includes a target detector for detecting Cordyceps sinensis, and the second-level regional target re-determination module includes a determiner for determining Cordyceps sinensis; the alarm module is connected to the computing chip, and the power supply module is connected to the circuit mainboard.

[0008] As a preferred embodiment, the alarm module is composed of one or more of a buzzer and an indicator light, and the power supply module is composed of one or more of a battery and a solar charging panel.

[0009] As a preferred embodiment, a fixing device for fixing the imaging device is provided on the imaging device, and the core control and calculation module is arranged in the fixing device.

[0010] As a preferred embodiment, the imaging device is provided with a stabilizing device for reducing shaking of the imaging device.

[0011] As a preferred embodiment, the imaging device is provided with a measuring bracket for maintaining a constant angle and distance between the imaging device and the ground surface.

[0012] A method for identifying Cordyceps sinensis targets in the wild, comprising the following steps:

[0013] S01: The imaging device is used to continuously scan the ground surface, and the imaging system of the imaging device collects scene images in the field of view in real time;

[0014] S02: The imaging device transmits the captured scene image to the core control and calculation module through the transmission line;

[0015] S03: The core control and calculation module detects and analyzes the scene image. If the detected scene image contains Cordyceps sinensis, step S04 is performed; otherwise, the next scene image obtained from step S01 is processed.

[0016] S04: If it is determined that the scene image contains the Cordyceps sinensis target, the warning module is triggered.

[0017] As a preferred method, the detection process of the core control and calculation module is as follows:

[0018] S201: After receiving a scene image, perform conventional image preprocessing on the scene image and output a preprocessed image;

[0019] S202: Lowering the resolution of the pre-processed image to output a low-resolution image;

[0020] S203: Using an image target detection algorithm on the low-resolution image, outputting one or more target candidate regions where Cordyceps sinensis is present;

[0021] S204: Based on the target candidate area, a high-resolution image corresponding to the target candidate frame is captured from the pre-processed image;

[0022] S205: Determine whether Cordyceps sinensis exists based on the high-resolution image.

[0023] As a preferred method, a method for identifying Cordyceps sinensis in the wild includes a training method for identifying Cordyceps sinensis, which includes the following steps:

[0024] S301: collecting an initial image containing Cordyceps sinensis;

[0025] S302: calibrating the coordinates of two vertices of the rectangular area where the Cordyceps sinensis is located in the initial image and outputting a preliminary processed image;

[0026] S303: performing image processing on the primary processed image to generate a secondary processed image, calculating calibration coordinate information of the Cordyceps sinensis in the secondary processed image, and outputting the secondary processed image and the calibration coordinate information of the Cordyceps sinensis in the secondary processed image;

[0027] S304: adjusting the image resolution of the secondary processed image and outputting a low-resolution secondary processed image, and recalculating the calibration coordinate information of the Cordyceps sinensis according to the ratio of the resolution adjustment and outputting the calibration coordinate information;

[0028] S305: Using the low-resolution secondary processed image and the calibration coordinate information to train the first-level object detection module to obtain object detector parameters;

[0029] S306: Based on the secondary processed image and the calibrated coordinate information of the Cordyceps sinensis in the secondary processed image, a rectangular image region containing the Cordyceps sinensis is cropped from the secondary processed image and data augmented to construct a positive sample. A rectangular image region of background weeds, three times the number of the positive sample target, is randomly cropped from the secondary processed image to construct a negative sample. The positive and negative sample rectangular regions are scaled to the same size.

[0030] S307: Using positive and negative samples to train the second-level regional target re-determination module to obtain the determiner parameters.

[0031] As a preferred approach, the first-level target detection module adopts a deep neural network, and the second-level regional target re-determination module adopts a convolutional neural network.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] (1) The present invention uses an imaging device to transmit surface video images collected by the imaging device to the core control and computing module for analysis, and uses an alarm device to sound an alarm for Cordyceps sinensis in the video image, which can quickly and effectively find Cordyceps sinensis and reduce the difficulty of manual search for Cordyceps sinensis;

[0034] (2) The present invention reduces the shaking of the imaging device during use by using a stabilizing device, thereby enhancing the stability of the imaging device and making the video image clearer;

[0035] (3) The present invention maintains the angle and distance between the imaging device and the ground surface by measuring the bracket, and the measuring bracket is in contact with the ground surface, which is conducive to determining the range of the imaging field of view; BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a structural schematic diagram of the present invention.

[0037] Figure 2 This is the detection flow chart of the core control and calculation module.

[0038] Figure 3 Flowchart of the training method for the first-level target detection module and the second-level target detection module.

[0039] Among them, 1 imaging device, 2 stabilization device, 3 fixing device, 4 measuring bracket, 5 transmission line, 6 core control and calculation module, 7 alarm module, 8 power supply module, 601 image preprocessing module, 602 image resolution adjustment module, 603 first-level target detection module, 604 candidate area high-resolution image module, 605 second-level area target re-determination module, 701 manual calibration training data module, 702 data amplification module, 703 image resolution adjustment module, 704 high-resolution target and background module. DETAILED DESCRIPTION

[0040] The present invention will be further described below with reference to the accompanying drawings. Embodiments of the present invention include but are not limited to the following examples.

[0041] Example 1:

[0042] See also Figure 1 A cordyceps sinensis field target recognition device includes an imaging device 1 for collecting surface scene images and a core control and computing module 6 for controlling various components and analyzing scene images. The imaging device 1 and the core control and computing module 6 are electrically connected via a transmission line 5. The core control and computing module 6 is connected to an alarm module 7 for warning of the presence of cordyceps sinensis and a power supply module 8 for providing power to the device.

[0043] In this embodiment, the imaging device 1 is used to collect surface video images, which are transmitted to the core control and computing module 6 via the transmission line 5. The core control and computing module 6 is used to control the electronic control of each component and the computational analysis of the video data, thereby analyzing whether Cordyceps sinensis is present within the imaging field of view of the imaging device 1. When the presence of Cordyceps sinensis in the imaging field of view is detected by the core control and computing module 6, the alarm module 7 issues a warning message. The power supply module 8 is used to supply power to the imaging device 1, the core control and computing module 6, and the alarm module 7, thereby reducing the difficulty of manual search for Cordyceps sinensis.

[0044] The core control and computing module 6 includes a circuit mainboard, a computing acceleration module, a computing chip and a storage device. The computing chip includes a first-level target detection module 603 for outputting one or more candidate target areas where Cordyceps sinensis exists and a second-level regional target re-determination module 605 for judging whether Cordyceps sinensis exists in the image based on the image. The first-level target detection module 603 includes a target detector for detecting Cordyceps sinensis, and the second-level regional target re-determination module 605 includes a determiner for determining Cordyceps sinensis; the alarm module 7 is connected to the computing chip, and the power supply module 8 is connected to the circuit mainboard.

[0045] The circuit board uses an Up Squared board or Raspberry Pi, and the computing acceleration module uses a computing acceleration chip. The computing acceleration chip compresses the trained model parameters in the core control and computing modules and then loads the compressed model and parameters into the Movidius hardware. The Movidius hardware has a dedicated chip VPU (Vector Processing Unit) that can accelerate the target detection deduction speed of the core control and computing modules. The computing chip uses the CPU N4200 or BCM2837, and the storage device uses 16GB of memory, a 256GB hard drive, or a 64GTF card. The alarm module 7 is composed of one or more of a buzzer and an indicator light, and the power supply module 8 is composed of one or more of a battery and a solar charging panel. The power supply module 8 supplies power to components such as the imaging device 1, the computing chip, the storage device, and the alarm module 7 via the transmission line 5.

[0046] Imaging device 1 is provided with a fixing device 3 for securing imaging device 1. Core control and computing module 6 is housed within fixing device 3. Fixing device 3 can be any of a handheld pole, a drone, and a robot dog. Fixing device 3 is used to secure imaging device 1 and protect core control and computing module 6. Specifically, any of these can be used. A handheld pole allows the operator to keep imaging device 1 close to the ground without bending their back. A drone can reduce damage to the grass during Cordyceps search. A robot dog allows the operator to avoid holding imaging device 1 and can move quickly through alpine grasslands, facilitating faster searches for Cordyceps.

[0047] The imaging device 1 is provided with a stabilizing device 2 for reducing the shaking of the imaging device 1. Since the imaging device 1 is prone to shaking when searching on alpine grasslands, resulting in unclear image quality, the stabilizing device 2 is helpful in reducing the shaking of the imaging device 1 and making the image quality clear.

[0048] The imaging device 1 is provided with a measuring bracket 4 for maintaining a constant angle and distance between the imaging device 1 and the ground. When the fixing device 3 for fixing the imaging device 1 is a handheld rod, the fixed measuring bracket 4 needs to be in contact with the ground so that the measuring bracket 4 can maintain a certain distance and angle with the ground, which is conducive to determining the imaging field of view of the imaging device 1; when the fixing device 3 for fixing the imaging device 1 is a drone or a robot dog, the drone always maintains a horizontal posture in flight and the robot dog always moves on the ground. The fixed bracket does not need to be in contact with the ground. It is only necessary to adjust the angle of the imaging device 1 on the fixing device 3 so that the imaging device 1 maintains a certain distance and angle with the ground, thereby determining the imaging field of view of the imaging device 1.

[0049] Example 2:

[0050] This embodiment is a specific application method of embodiment 1.

[0051] See also Figure 1 A method for identifying a Cordyceps sinensis target in the wild comprises the following steps:

[0052] S01: The imaging device 1 is used to continuously scan the ground surface, and the imaging system of the imaging device 1 collects and acquires scene images in the field of view in real time;

[0053] S02: The imaging device 1 transmits the captured scene image to the core control and calculation module via the transmission line 5;

[0054] S03: The core control and calculation module 6 detects and analyzes the scene image. If the detected scene image contains Cordyceps sinensis, step S04 is performed; otherwise, the next scene image obtained from step S01 is processed.

[0055] S04: If it is determined that the scene image contains the Cordyceps sinensis target, the warning module is triggered.

[0056] The imaging system is usually color imaging, but can be expanded to use depth of field, infrared, multispectral, hyperspectral and other imaging information.

[0057] See also Figure 2 The detection process of the core control and calculation module 6 is as follows:

[0058] S201: After receiving a scene image, perform conventional image preprocessing on the scene image and output a preprocessed image;

[0059] S202: Lowering the resolution of the pre-processed image to output a low-resolution image;

[0060] S203: Using an image target detection algorithm on the low-resolution image, outputting one or more target candidate regions where Cordyceps sinensis is present;

[0061] S204: Based on the target candidate area, a high-resolution image corresponding to the target candidate frame is captured from the pre-processed image;

[0062] S205: Determine whether Cordyceps sinensis exists based on the high-resolution image.

[0063] The image preprocessing module 601 performs scene image preprocessing and outputs a command for the preprocessed image. The preprocessing includes but is not limited to operations such as automatic brightness adjustment and color adjustment. The image resolution provided by the conventional imaging device 1 is relatively high. The image resolution adjustment module 602 executes a command to lower the resolution of the preprocessed image and output a low-resolution image, which is conducive to high-speed measurement of Cordyceps sinensis. The first-level target detection module 603 executes a command to output one or more target candidate areas where Cordyceps sinensis exists from the low-resolution image using a target detection algorithm. Target detection is to determine the position and category of a specific target in the input scene image. The candidate area high-resolution image module 604 executes a command to increase the resolution of the target candidate area and output a high-resolution image. The high-resolution image has a higher resolution than the low-resolution image and can be the same as the resolution of the preprocessed image. The second-level area target re-determination module 605 executes a command to determine whether Cordyceps sinensis exists in the high-resolution image, and again determines whether the area is a Cordyceps sinensis target.

[0064] See also Figure 3 A method for identifying Cordyceps sinensis in the wild includes a training method for identifying Cordyceps sinensis, comprising the following steps:

[0065] S301: collecting an initial image containing Cordyceps sinensis;

[0066] S302: calibrating the coordinates of two vertices of the rectangular area where the Cordyceps sinensis is located in the initial image and outputting a preliminary processed image;

[0067] S303: performing image processing on the primary processed image to generate a secondary processed image, calculating calibration coordinate information of the Cordyceps sinensis in the secondary processed image, and outputting the secondary processed image and the calibration coordinate information of the Cordyceps sinensis in the secondary processed image;

[0068] S304: adjusting the image resolution of the secondary processed image and outputting a low-resolution secondary processed image, and recalculating the calibration coordinate information of the Cordyceps sinensis according to the ratio of the resolution adjustment and outputting the calibration coordinate information;

[0069] S305: Train the first-level target detection module 603 using the low-resolution secondary processed image and the calibration coordinate information to obtain target detector parameters;

[0070] S306: Based on the secondary processed image and the calibrated coordinate information of the Cordyceps sinensis in the secondary processed image, a rectangular image region containing the Cordyceps sinensis is cropped from the secondary processed image and data augmented to construct a positive sample. A rectangular image region of background weeds, three times the number of the positive sample target, is randomly cropped from the secondary processed image to construct a negative sample. The positive and negative sample rectangular regions are scaled to the same size.

[0071] S307: Use positive and negative samples to train the second-level regional target re-determination module 605 to obtain the determiner parameters.

[0072] The manual calibration training data module 701 performs coordinate calibration on the two vertex coordinates of the rectangular area where the Cordyceps sinensis is located in the initial image and outputs a preliminary processed image. The two vertex coordinates are specifically the coordinates of the upper left and lower right points of the rectangular area. This step is to output the initial processed image with the coordinates of the upper left and lower right points of the rectangular area where the Cordyceps sinensis is located calibrated; the data augmentation module 702 performs image processing on the preliminary processed image and calculates the calibration coordinate information of the Cordyceps sinensis, outputs a secondary processed image and a command for the calibration coordinate information of the Cordyceps sinensis in the secondary processed image, wherein the image processing is processing by mirroring, rotating, scaling, brightness, and hue adjustment on the preliminary processed image; the image resolution adjustment module 602 performs lowering the image resolution of the secondary processed image to output a low-resolution secondary processed image, and recalculates the calibration coordinate information of the Cordyceps sinensis according to the adjustment of its resolution, and outputs a command for the calibration coordinate information; the first-level target detection module 603 is trained based on the low-resolution secondary processed image and the calibration coordinate information to obtain the target detector parameters. The high-resolution target and background module 704 executes the operation of cropping the rectangular image area where the Cordyceps sinensis is located in the secondary processed image and performing data enhancement to construct a positive sample. It randomly crops the rectangular image area of ​​background weeds three times the number of positive sample targets from the secondary processed image to construct a negative sample. The positive and negative sample rectangular areas are scaled to the same size to construct a command for positive and negative samples. The second-level regional target re-determination module 605 is trained based on the positive and negative samples to obtain the determiner parameters.

[0073] The first-level target detection module 603 uses a deep neural network, while the second-level area target re-determination module 605 uses a convolutional neural network. The second-level area target re-determination module 605 is specifically designed as a two-class classifier based on a convolutional neural network. The deep neural network is composed of multiple layers of convolutional neural networks, and the convolution kernel size of each layer and the total number of network layers must be determined. The deep neural network uses a Movidius neural computing chip.

[0074] The above are embodiments of the present invention. The above embodiments and the specific parameters therein are only for the purpose of clearly describing the verification process of the invention and are not intended to limit the scope of patent protection of the present invention. The scope of patent protection of the present invention shall still be subject to the claims. Any equivalent structural changes made by using the contents of the description and drawings of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A Cordyceps sinensis field target identification device, characterized by: The invention comprises an imaging device (1) for collecting surface scene images and a core control and calculation module (6) for controlling various components and analyzing scene images, wherein the imaging device (1) and the core control and calculation module (6) are electrically connected via a transmission line (5), and the core control and calculation module (6) is connected to an alarm module (7) for warning of the presence of cordyceps sinensis and a power supply module (8) for providing power to the device; The core control and calculation module (6) includes a circuit mainboard, a calculation acceleration module, a calculation chip and a storage device. The calculation chip includes a first-level target detection module (603) for outputting one or more candidate regions where Cordyceps sinensis exists and a second-level region target re-determination module (605) for determining whether Cordyceps sinensis exists in a scene image. The first-level target detection module (603) includes a target detector for detecting Cordyceps sinensis, and the second-level region target re-determination module (605) includes a determiner for determining whether Cordyceps sinensis exists. The alarm module (7) is connected to the calculation chip, and the power supply module (8) is connected to the circuit mainboard. The alarm module (7) is composed of one or more of a buzzer and an indicator light, and the power supply module (8) is composed of one or more of a battery and a solar charging panel; The imaging device (1) is provided with a fixing device (3) for fixing the imaging device (1), and the core control and calculation module (6) is arranged in the fixing device (3); The imaging device (1) is provided with a stabilizing device (2) for reducing shaking of the imaging device (1); The imaging device (1) is provided with a measuring bracket (4) for maintaining a constant angle and distance between the imaging device (1) and the ground surface; The Cordyceps sinensis field target recognition device is used to recognize Cordyceps sinensis field targets. The Cordyceps sinensis field target recognition method comprises the following steps: S01: The imaging device (1) is held to continuously scan the ground surface, and the imaging system of the imaging device (1) collects and obtains scene images in the field of view in real time; S02: The imaging device (1) transmits the captured scene image to the core control and calculation module (6) via the transmission line (5); S03: The core control and calculation module (6) detects and analyzes the scene image. If the detected scene image contains Cordyceps sinensis, step S04 is performed; otherwise, the next scene image obtained from step S01 is processed. S04: If it is determined that the scene image contains the Cordyceps sinensis target, triggering the warning module; The detection process of the core control and calculation module (6) includes the following steps: S201: After receiving a scene image, perform conventional image preprocessing on the scene image and output a preprocessed image; S202: Lowering the resolution of the pre-processed image to output a low-resolution image; S203: Using an image target detection algorithm on the low-resolution image, outputting one or more target candidate regions where Cordyceps sinensis is present; S204: Based on the target candidate area, a high-resolution image corresponding to the target candidate frame is captured from the pre-processed image; S205: Determine whether Cordyceps sinensis exists based on the high-resolution image; The method for identifying Cordyceps sinensis in the wild also includes a training method for identifying Cordyceps sinensis, which includes the following steps: S301: collecting an initial image containing Cordyceps sinensis; S302: calibrating the coordinates of two vertices of the rectangular area where the Cordyceps sinensis is located in the initial image and outputting a preliminary processed image; S303: performing image processing on the primary processed image to generate a secondary processed image, calculating calibration coordinate information of the Cordyceps sinensis in the secondary processed image, and outputting the secondary processed image and the calibration coordinate information of the Cordyceps sinensis in the secondary processed image; S304: adjusting the image resolution of the secondary processed image and outputting a low-resolution secondary processed image, and recalculating the calibration coordinate information of the Cordyceps sinensis according to the ratio of the resolution adjustment and outputting the calibration coordinate information; S305: using the low-resolution secondary processed image and the calibration coordinate information to train the first-level target detection module (603) to obtain target detector parameters; S306: Based on the secondary processed image and the calibrated coordinate information of the Cordyceps sinensis in the secondary processed image, a rectangular image region containing the Cordyceps sinensis is cropped from the secondary processed image and data augmented to construct a positive sample. A rectangular image region of background weeds, three times the number of the positive sample target, is randomly cropped from the secondary processed image to construct a negative sample. The positive and negative sample rectangular regions are scaled to the same size. S307: Using positive and negative samples, the second-level regional target re-determination module (605) is trained to obtain the determiner parameters.

2. A method for identifying Cordyceps sinensis in the field, characterized by: The method of using the Cordyceps sinensis field target identification device as claimed in claim 1 to identify the target comprises the following steps: S01: The imaging device (1) is held to continuously scan the ground surface, and the imaging system of the imaging device (1) collects and obtains scene images in the field of view in real time; S02: The imaging device (1) transmits the captured scene image to the core control and calculation module (6) via the transmission line (5); S03: The core control and calculation module (6) detects and analyzes the scene image. If the detected scene image contains Cordyceps sinensis, step S04 is performed; otherwise, the next scene image obtained from step S01 is processed. S04: If it is determined that the scene image contains the Cordyceps sinensis target, triggering the warning module; The detection process of the core control and calculation module (6) includes the following steps: S201: After receiving a scene image, perform conventional image preprocessing on the scene image and output a preprocessed image; S202: Lowering the resolution of the pre-processed image to output a low-resolution image; S203: Using an image target detection algorithm on the low-resolution image, outputting one or more target candidate regions where Cordyceps sinensis is present; S204: Based on the target candidate area, a high-resolution image corresponding to the target candidate frame is captured from the pre-processed image; S205: Determine whether Cordyceps sinensis exists based on the high-resolution image; The method for identifying Cordyceps sinensis in the wild also includes a training method for identifying Cordyceps sinensis, which includes the following steps: S301: collecting an initial image containing Cordyceps sinensis; S302: calibrating the coordinates of two vertices of the rectangular area where the Cordyceps sinensis is located in the initial image and outputting a preliminary processed image; S303: performing image processing on the primary processed image to generate a secondary processed image, calculating calibration coordinate information of the Cordyceps sinensis in the secondary processed image, and outputting the secondary processed image and the calibration coordinate information of the Cordyceps sinensis in the secondary processed image; S304: adjusting the image resolution of the secondary processed image and outputting a low-resolution secondary processed image, and recalculating the calibration coordinate information of the Cordyceps sinensis according to the ratio of the resolution adjustment and outputting the calibration coordinate information; S305: using the low-resolution secondary processed image and the calibration coordinate information to train the first-level target detection module (603) to obtain target detector parameters; S306: Based on the secondary processed image and the calibrated coordinate information of the Cordyceps sinensis in the secondary processed image, a rectangular image region containing the Cordyceps sinensis is cropped from the secondary processed image and data augmented to construct a positive sample. A rectangular image region of background weeds, three times the number of the positive sample target, is randomly cropped from the secondary processed image to construct a negative sample. The positive and negative sample rectangular regions are scaled to the same size. S307: Using positive and negative samples, the second-level regional target re-determination module (605) is trained to obtain the determiner parameters.

3. The method for identifying Cordyceps sinensis in the field according to claim 2, wherein: The first-level target detection module (603) adopts a deep neural network, and the second-level area target re-determination module (605) adopts a convolutional neural network.

Images