Soybean abnormal plant identification method based on unmanned remote sensing and image identification

Abstract

The invention discloses a soybean abnormal plant identification method based on unmanned remote sensing and image identification, and relates to the field of remote sensing and image depth processing. Comprising the following steps: unmanned aerial vehicle path planning: carrying an RGB camera on an unmanned aerial vehicle, and planning the path of the unmanned aerial vehicle according to an area needing image acquisition; image acquisition and preprocessing: in the flight process of the unmanned aerial vehicle, performing low-altitude remote sensing shooting on soybean images of the soybean field by an RGB camera to obtain image data, and then preprocessing the image data; and image background segmentation and optimization processing: carrying out further processing, feature extraction and identification of image background segmentation and optimization on the image, carrying out feature extraction on the processed image, and then carrying out abnormal plant identification according to a feature extraction result. Through the advantages of the unmanned aerial vehicle remote sensing and image recognition technology, precise recognition and processing of abnormal plants in the soybean field are achieved, positioning of the unmanned aerial vehicle is optimized, and the precision and comprehensiveness of image collection can be guaranteed.

Description

Technical Field

[0001] The invention relates to the field of remote sensing and image depth processing, and in particular to a soybean offspring identification method based on unmanned remote sensing and image recognition. Background Art

[0002] Originally produced in China, soybeans are cultivated in various parts of China and are also widely cultivated around the world. Soybeans are one of China's important food crops, with a 5,000-year history of cultivation. They were called "shu" in ancient times and are mainly produced in Northeast China. Soybeans are a crop whose seeds are rich in plant protein, with a protein content of 35% to 40%. Soybeans are most commonly used to make various soy products, extract soybean oil, brew soy sauce and extract protein.

[0003] In the process of soybean seed production, the soybean seeds need to be screened first, and then the screened soybean seeds are planted and cultivated. During the planting and cultivation process, the impure plants are identified and removed, and then the pure plants are retained to obtain soybean seeds. After repeating the above method many times, soybean seeds with higher purity are obtained.

[0004] In the prior art, the screening of foreign plants still relies on manual work, which not only has low screening efficiency but also low screening accuracy.

[0005] To this end, the present invention proposes a soybean offspring identification method based on unmanned remote sensing and image recognition Summary of the invention

[0006] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a soybean offspring identification method based on unmanned remote sensing and image recognition.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A soybean offspring identification method based on unmanned remote sensing and image recognition comprises the following steps:

[0009] S1: UAV path planning, the RGB camera is mounted on the UAV, and the UAV path is planned according to the area where the image acquisition is required;

[0010] S2: Image acquisition and preprocessing. During the flight, the RGB camera of the UAV takes low-altitude remote sensing photos of soybeans in the soybean field, obtains image data, and then preprocesses the image data;

[0011] S3: image background segmentation and optimization processing, further processing of image background segmentation and optimization;

[0012] S4: Feature extraction and recognition: extract features from the processed image, and then identify the foreign plants based on the feature extraction results.

[0013] Preferably: the step S1 includes the following steps:

[0014] S11: Setting up multiple drones for collaborative operation and inputting the drone flight mission and path planning information into the control system of each drone;

[0015] S12: The Beidou positioning module built into the drone obtains the flight position of the drone in real time during flight, and controls the drone to make corrections according to the planned path if there is any deviation;

[0016] S13: The signal strength detection unit in the Beidou positioning module detects the positioning signal strength in real time;

[0017] S14: setting a lower limit threshold of the signal strength according to the positioning deviation caused by the signal strength and the required positioning deviation accuracy;

[0018] S15: When the signal strength of the positioning signal is lower than the threshold, a mutual voting-based algorithm is used to perform positioning using the networking of drones.

[0019] Preferably: the step S15 comprises the following steps:

[0020] S151: Obtain theoretical positions between each UAV according to path planning;

[0021] S152: judging the actual positions of the drones according to the visual markers collected by the visual collection module;

[0022] S153: Using the deviation between the theoretical position and the actual position, voting on the position accuracy of the target UAV, and determining the reference UAV according to the voting result;

[0023] S154: The visual acquisition module acquires the visual markers on the reference UAV and calculates the orientation, and adjusts the orientation of the UAV.

[0024] Preferably: the step S153 comprises the following steps:

[0025] S1531: Setting the allowable deviation coefficient k;

[0026] S1532: Decompose the orientation along the three-dimensional space to obtain the theoretical distance L of the X, Y, and Z axes respectively X , L Y , L Z And the actual distance L′ X , L′ Y , L′ Z ;

[0027] S1533: Determination ①: L′ X ∈[(1-k)LX , (1+k)L X ], ②: L′ Y ∈[(1-k)L Y , (1+k)L Y ]、③L′ X ∈[(1-k)L Z , (1+k)L Z ];

[0028] S1534: Determine the voting score based on the conditions of ①②③;

[0029] S1535: Take the average of the voting results of all drones on the target drone to obtain the voting score of the target drone.

[0030] Preferably: in the step S1531, k∈[0.03, 0.10], in the step S1534, if ①②③ all meet the requirements, the voting result for the target drone is 1; if one of ①②③ does not meet the requirements, the voting result for the target drone is 2 / 3; if two of ①②③ do not meet the requirements, the voting result for the target drone is 1 / 3; if none of ①②③ meet the requirements, the voting result is 0.

[0031] Preferably: in the step S2, the drone is a quad-rotor drone, the resolution of the RGB camera is 40 million pixels, and the flight altitude of the drone is 3.5-4.5 meters.

[0032] Preferably: in the step S2, the image preprocessing comprises the following steps:

[0033] S21: grayscale processing of the image, using the weighted average method for grayscale conversion, the grayscale conversion formula is Gray = 0.299*R + 0.587*G + 0.114*B, in order to reduce storage space and simplify calculation;

[0034] S22: De-noising, using extreme adaptive median filtering to remove salt and pepper noise in the image, and then using wavelet filtering to remove Gaussian noise;

[0035] S23: Image enhancement, using the formula: l′(x, y)=log[l(x, y) / F(x, y)] to increase the image, where l(x, y) is the original image, F(x, y) is the illumination component, and l′(x, y) is the enhanced image.

[0036] Preferably: the step S3 comprises the following steps:

[0037] S31: using the histogram of the calculated image, calculating the intra-class variance of the foreground and background, and finding the threshold point that maximizes the inter-class variance as the segmentation point;

[0038] S32: Perform erosion and dilation operations on the segmented foreground image to remove small noise points and fill holes.

[0039] Preferably: the step S4 comprises the following steps:

[0040] S41: feature selection, extracting soybean leaf shape features, texture features and color features;

[0041] S42: Model construction, using lightweight convolutional neural network for feature learning and classification; the network structure includes convolution layer, depthwise separable convolution layer, batch normalization layer and ReLU activation function, as well as fully connected layer and Softmax classifier;

[0042] S43: Model training, using labeled soybean and weed image datasets for training, using cross-validation and early stopping strategies to prevent overfitting, while using learning rate decay and momentum to optimize the training process;

[0043] S44: Identification of foreign plants: input the processed image into the model, the model identifies foreign plants in the image, and outputs the identification result.

[0044] Preferably: in the step S41, the leaf shape features include area, perimeter and aspect ratio, the texture features are gray-level co-occurrence matrix or local binary value, and the color features are RGB value or HSV value.

[0045] The beneficial effects of the present invention are:

[0046] 1. The advantages of the drone remote sensing and image recognition technology of the present invention realize the accurate identification and treatment of weeds in soybean fields. Compared with the traditional manual identification method, this method has the advantages of high efficiency, high accuracy and wide application range.

[0047] 2. The present invention optimizes the positioning of the UAV, thereby ensuring the reliability of the position of the UAV, thereby ensuring the accuracy and comprehensiveness of image acquisition. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flow chart of a soybean offal identification method based on unmanned remote sensing and image recognition proposed by the present invention;

[0049] Figure 2 A flow chart of unmanned aerial vehicle path planning for a soybean offspring identification method based on unmanned remote sensing and image recognition proposed by the present invention;

[0050] Figure 3 This is a flowchart of image acquisition and preprocessing of a soybean offspring identification method based on unmanned remote sensing and image recognition proposed by the present invention;

[0051] Figure 4 A flowchart of image background segmentation and optimization processing of a soybean offspring identification method based on unmanned remote sensing and image recognition proposed by the present invention;

[0052] Figure 5 The present invention provides a feature extraction and recognition flow chart of a soybean offspring recognition method based on unmanned remote sensing and image recognition. DETAILED DESCRIPTION

[0053] The technical solution of the present invention is further described in detail below in conjunction with specific implementation methods.

[0054] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", and "set" should be understood in a broad sense, for example, they can be fixedly connected or set, or detachably connected or set, or integrally connected or set. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0055] Embodiment 1:

[0056] A soybean offspring identification method based on unmanned remote sensing and image recognition comprises the following steps:

[0057] S1: UAV path planning, the RGB camera is mounted on the UAV, and the UAV path is planned according to the area where the image acquisition is required;

[0058] S2: Image acquisition and preprocessing. During the flight, the RGB camera of the UAV takes low-altitude remote sensing photos of soybeans in the soybean field, obtains image data, and then preprocesses the image data;

[0059] S3: image background segmentation and optimization processing, further processing of image background segmentation and optimization;

[0060] S4: Feature extraction and recognition: extract features from the processed image, and then identify the foreign plants based on the feature extraction results.

[0061] Embodiment 2:

[0062] A soybean offspring identification method based on unmanned remote sensing and image recognition comprises the following steps:

[0063] S1: UAV path planning, the RGB camera is mounted on the UAV, and the UAV path is planned according to the area where the image acquisition is required;

[0064] S2: Image acquisition and preprocessing. During the flight, the RGB camera of the UAV takes low-altitude remote sensing photos of soybeans in the soybean field, obtains image data, and then preprocesses the image data;

[0065] S3: image background segmentation and optimization processing, further processing of image background segmentation and optimization;

[0066] S4: Feature extraction and recognition: extract features from the processed image, and then identify the foreign plants based on the feature extraction results.

[0067] The step S1 includes the following steps:

[0068] S11: Setting up multiple drones for collaborative operation and inputting the drone flight mission and path planning information into the control system of each drone;

[0069] S12: The Beidou positioning module built into the drone obtains the flight position of the drone in real time during flight, and controls the drone to make corrections according to the planned path if there is any deviation;

[0070] S13: The signal strength detection unit in the Beidou positioning module detects the positioning signal strength in real time;

[0071] S14: setting a lower limit threshold of the signal strength according to the positioning deviation caused by the signal strength and the required positioning deviation accuracy;

[0072] S15: When the signal strength of the positioning signal is lower than the threshold, a mutual voting-based algorithm is used to perform positioning using the networking of drones.

[0073] Embodiment 3:

[0074] A soybean offspring identification method based on unmanned remote sensing and image recognition comprises the following steps:

[0075] S1: UAV path planning, the RGB camera is mounted on the UAV, and the UAV path is planned according to the area where the image acquisition is required;

[0076] S2: Image acquisition and preprocessing. During the flight, the RGB camera of the UAV takes low-altitude remote sensing photos of soybeans in the soybean field, obtains image data, and then preprocesses the image data;

[0077] S3: image background segmentation and optimization processing, further processing of image background segmentation and optimization;

[0078] S4: Feature extraction and recognition: extract features from the processed image, and then identify the foreign plants based on the feature extraction results.

[0079] The step S1 includes the following steps:

[0080] S11: Setting up multiple drones for collaborative operation and inputting the drone flight mission and path planning information into the control system of each drone;

[0081] S12: The Beidou positioning module built into the drone obtains the flight position of the drone in real time during flight, and controls the drone to make corrections according to the planned path if there is any deviation;

[0082] S13: The signal strength detection unit in the Beidou positioning module detects the positioning signal strength in real time;

[0083] S14: setting a lower limit threshold of the signal strength according to the positioning deviation caused by the signal strength and the required positioning deviation accuracy;

[0084] S15: When the signal strength of the positioning signal is lower than the threshold, a mutual voting-based algorithm is used to perform positioning using the networking of drones.

[0085] The step S15 comprises the following steps:

[0086] S151: Obtain theoretical positions between each UAV according to path planning;

[0087] S152: judging the actual positions of the drones according to the visual markers collected by the visual collection module;

[0088] S153: Using the deviation between the theoretical position and the actual position, voting on the position accuracy of the target UAV, and determining the reference UAV according to the voting result;

[0089] S154: The visual acquisition module acquires the visual markers on the reference UAV and calculates the orientation, and adjusts the orientation of the UAV.

[0090] Embodiment 4:

[0091] A soybean offspring identification method based on unmanned remote sensing and image recognition comprises the following steps:

[0092] S1: UAV path planning, the RGB camera is mounted on the UAV, and the UAV path is planned according to the area where the image acquisition is required;

[0093] S2: Image acquisition and preprocessing. During the flight, the RGB camera of the UAV takes low-altitude remote sensing photos of soybeans in the soybean field, obtains image data, and then preprocesses the image data;

[0094] S3: image background segmentation and optimization processing, further processing of image background segmentation and optimization;

[0095] S4: Feature extraction and recognition: extract features from the processed image, and then identify the foreign plants based on the feature extraction results.

[0096] The step S1 includes the following steps:

[0097] S11: Setting up multiple drones for collaborative operation and inputting the drone flight mission and path planning information into the control system of each drone;

[0098] S12: The Beidou positioning module built into the drone obtains the flight position of the drone in real time during flight, and controls the drone to make corrections according to the planned path if there is any deviation;

[0099] S13: The signal strength detection unit in the Beidou positioning module detects the positioning signal strength in real time;

[0100] S14: setting a lower limit threshold of the signal strength according to the positioning deviation caused by the signal strength and the required positioning deviation accuracy;

[0101] S15: When the signal strength of the positioning signal is lower than the threshold, a mutual voting-based algorithm is used to perform positioning using the networking of drones.

[0102] The step S15 comprises the following steps:

[0103] S151: Obtain theoretical positions between each UAV according to path planning;

[0104] S152: judging the actual positions of the drones according to the visual markers collected by the visual collection module;

[0105] S153: Using the deviation between the theoretical position and the actual position, voting on the position accuracy of the target UAV, and determining the reference UAV according to the voting result;

[0106] S154: The visual acquisition module acquires the visual markers on the reference UAV and calculates the orientation, and adjusts the orientation of the UAV.

[0107] The step S153 comprises the following steps:

[0108] S1531: Setting the allowable deviation coefficient k;

[0109] S1532: Decompose the orientation along the three-dimensional space to obtain the theoretical distance L of the X, Y, and Z axes respectively X , L Y , L Z And the actual distance L′ X , L′ Y , L′Z ;

[0110] S1533: Determination ①: L′ X ∈[(1-k)L X , (1+k)L X ], ②: L′ Y ∈[(1-k)L Y , (1+k)L Y ]、③L′ X ∈[(1-k)L Z , (1+k)L Z ];

[0111] S1534: Determine the voting score based on the conditions of ①②③;

[0112] S1535: Take the average of the voting results of all drones on the target drone to obtain the voting score of the target drone.

[0113] Embodiment 5:

[0114] A soybean offspring identification method based on unmanned remote sensing and image recognition comprises the following steps:

[0115] S1: UAV path planning, the RGB camera is mounted on the UAV, and the UAV path is planned according to the area where the image acquisition is required;

[0116] S2: Image acquisition and preprocessing. During the flight, the RGB camera of the UAV takes low-altitude remote sensing photos of soybeans in the soybean field, obtains image data, and then preprocesses the image data;

[0117] S3: image background segmentation and optimization processing, further processing of image background segmentation and optimization;

[0118] S4: Feature extraction and recognition: extract features from the processed image, and then identify the foreign plants based on the feature extraction results.

[0119] The step S1 includes the following steps:

[0120] S11: Setting up multiple drones for collaborative operation and inputting the drone flight mission and path planning information into the control system of each drone;

[0121] S12: The Beidou positioning module built into the drone obtains the flight position of the drone in real time during flight, and controls the drone to make corrections according to the planned path if there is any deviation;

[0122] S13: The signal strength detection unit in the Beidou positioning module detects the positioning signal strength in real time;

[0123] S14: setting a lower limit threshold of the signal strength according to the positioning deviation caused by the signal strength and the required positioning deviation accuracy;

[0124] S15: When the signal strength of the positioning signal is lower than the threshold, a mutual voting-based algorithm is used to perform positioning using the networking of drones.

[0125] The step S15 comprises the following steps:

[0126] S151: Obtain theoretical positions between each UAV according to path planning;

[0127] S152: judging the actual positions of the drones according to the visual markers collected by the visual collection module;

[0128] S153: Using the deviation between the theoretical position and the actual position, voting on the position accuracy of the target UAV, and determining the reference UAV according to the voting result;

[0129] S154: The visual acquisition module acquires the visual markers on the reference UAV and calculates the orientation, and adjusts the orientation of the UAV.

[0130] The step S153 comprises the following steps:

[0131] S1531: Setting the allowable deviation coefficient k;

[0132] S1532: Decompose the orientation along the three-dimensional space to obtain the theoretical distance L of the X, Y, and Z axes respectively X , L Y , L Z And the actual distance L′ X , L′ Y , L′ Z ;

[0133] S1533: Determination ①: L′ X ∈[(1-k)L X , (1+k)L X ], ②: L′ Y ∈[(1-k)L Y , (1+k)L Y ]、③L′ X ∈[(1-k)L Z , (1+k)L Z ];

[0134] S1534: Determine the voting score based on the conditions of ①②③;

[0135] S1535: Take the average of the voting results of all drones on the target drone to obtain the voting score of the target drone.

[0136] In step S1531, k∈[0.03, 0.10], in step S1534, if ①②③ all meet the requirements, the voting result for the target drone is 1; if one of ①②③ does not meet the requirements, the voting result for the target drone is 2 / 3; if two of ①②③ do not meet the requirements, the voting result for the target drone is 1 / 3; if none of ①②③ meet the requirements, the voting result is 0.

[0137] Embodiment 6:

[0138] A soybean offspring identification method based on unmanned remote sensing and image recognition comprises the following steps:

[0139] S1: UAV path planning, the RGB camera is mounted on the UAV, and the UAV path is planned according to the area where the image acquisition is required;

[0140] S2: Image acquisition and preprocessing. During the flight, the RGB camera of the UAV takes low-altitude remote sensing photos of soybeans in the soybean field, obtains image data, and then preprocesses the image data;

[0141] S3: image background segmentation and optimization processing, further processing of image background segmentation and optimization;

[0142] S4: Feature extraction and recognition: extract features from the processed image, and then identify the foreign plants based on the feature extraction results.

[0143] The step S1 includes the following steps:

[0144] S11: Setting up multiple drones for collaborative operation and inputting the drone flight mission and path planning information into the control system of each drone;

[0145] S12: The Beidou positioning module built into the drone obtains the flight position of the drone in real time during flight, and controls the drone to make corrections according to the planned path if there is any deviation;

[0146] S13: The signal strength detection unit in the Beidou positioning module detects the positioning signal strength in real time;

[0147] S14: setting a lower limit threshold of the signal strength according to the positioning deviation caused by the signal strength and the required positioning deviation accuracy;

[0148] S15: When the signal strength of the positioning signal is lower than the threshold, a mutual voting-based algorithm is used to perform positioning using the networking of drones.

[0149] The step S15 comprises the following steps:

[0150] S151: Obtain theoretical positions between each UAV according to path planning;

[0151] S152: judging the actual positions of the drones according to the visual markers collected by the visual collection module;

[0152] S153: Using the deviation between the theoretical position and the actual position, voting on the position accuracy of the target UAV, and determining the reference UAV according to the voting result;

[0153] S154: The visual acquisition module acquires the visual markers on the reference UAV and calculates the orientation, and adjusts the orientation of the UAV.

[0154] The step S153 comprises the following steps:

[0155] S1531: Setting the allowable deviation coefficient k;

[0156] S1532: Decompose the orientation along the three-dimensional space to obtain the theoretical distance L of the X, Y, and Z axes respectively X , L Y , L Z And the actual distance L′ X , L′ Y , L′ Z ;

[0157] S1533: Determination ①: L′ X ∈[(1-k)L X , (1+k)L X ], ②: L′ Y ∈[(1-k)L Y , (1+k)L Y ]、③L′ X ∈[(1-k)L Z , (1+k)L Z ];

[0158] S1534: Determine the voting score based on the conditions of ①②③;

[0159] S1535: Take the average of the voting results of all drones on the target drone to obtain the voting score of the target drone.

[0160] In step S1531, k∈[0.03, 0.10], in step S1534, if ①②③ all meet the requirements, the voting result for the target drone is 1; if one of ①②③ does not meet the requirements, the voting result for the target drone is 2 / 3; if two of ①②③ do not meet the requirements, the voting result for the target drone is 1 / 3; if none of ①②③ meet the requirements, the voting result is 0.

[0161] In the step S2, the drone is a quad-rotor drone, the resolution of the RGB camera is 40 million pixels, and the flight altitude of the drone is 3.5-4.5 meters.

[0162] Embodiment 7:

[0163] A soybean offspring identification method based on unmanned remote sensing and image recognition comprises the following steps:

[0164] S1: UAV path planning, the RGB camera is mounted on the UAV, and the UAV path is planned according to the area where the image acquisition is required;

[0165] S2: Image acquisition and preprocessing. During the flight, the RGB camera of the UAV takes low-altitude remote sensing photos of soybeans in the soybean field, obtains image data, and then preprocesses the image data;

[0166] S3: image background segmentation and optimization processing, further processing of image background segmentation and optimization;

[0167] S4: Feature extraction and recognition: extract features from the processed image, and then identify the foreign plants based on the feature extraction results.

[0168] The step S1 includes the following steps:

[0169] S11: Setting up multiple drones for collaborative operation and inputting the drone flight mission and path planning information into the control system of each drone;

[0170] S12: The Beidou positioning module built into the drone obtains the flight position of the drone in real time during flight, and controls the drone to make corrections according to the planned path if there is any deviation;

[0171] S13: The signal strength detection unit in the Beidou positioning module detects the positioning signal strength in real time;

[0172] S14: setting a lower limit threshold of the signal strength according to the positioning deviation caused by the signal strength and the required positioning deviation accuracy;

[0173] S15: When the signal strength of the positioning signal is lower than the threshold, a mutual voting-based algorithm is used to perform positioning using the networking of drones.

[0174] The step S15 comprises the following steps:

[0175] S151: Obtain theoretical positions between each UAV according to path planning;

[0176] S152: judging the actual positions of the drones according to the visual markers collected by the visual collection module;

[0177] S153: Using the deviation between the theoretical position and the actual position, voting on the position accuracy of the target UAV, and determining the reference UAV according to the voting result;

[0178] S154: The visual acquisition module acquires the visual markers on the reference UAV and calculates the orientation, and adjusts the orientation of the UAV.

[0179] The step S153 comprises the following steps:

[0180] S1531: Setting the allowable deviation coefficient k;

[0181] S1532: Decompose the orientation along the three-dimensional space to obtain the theoretical distance L of the X, Y, and Z axes respectively X , L Y , L Z And the actual distance L′ X , L′ Y , L′ Z ;

[0182] S1533: Determination ①: L′ X ∈[(1-k)L X , (1+k)L X ], ②: L′ Y ∈[(1-k)L Y , (1+k)L Y ]、③L′ X ∈[(1-k)L Z , (1+k)L Z ];

[0183] S1534: Determine the voting score based on the conditions of ①②③;

[0184] S1535: Take the average of the voting results of all drones on the target drone to obtain the voting score of the target drone.

[0185] In step S1531, k∈[0.03, 0.10], in step S1534, if ①②③ all meet the requirements, the voting result for the target drone is 1; if one of ①②③ does not meet the requirements, the voting result for the target drone is 2 / 3; if two of ①②③ do not meet the requirements, the voting result for the target drone is 1 / 3; if none of ①②③ meet the requirements, the voting result is 0.

[0186] In the step S2, the drone is a quad-rotor drone, the resolution of the RGB camera is 40 million pixels, and the flight altitude of the drone is 3.5-4.5 meters.

[0187] In the step S2, the image preprocessing includes the following steps:

[0188] S21: grayscale processing of the image, using the weighted average method for grayscale conversion, the grayscale conversion formula is Gray = 0.299*R + 0.587*G + 0.114*B, in order to reduce storage space and simplify calculation;

[0189] S22: De-noising, using extreme adaptive median filtering to remove salt and pepper noise in the image, and then using wavelet filtering to remove Gaussian noise;

[0190] S23: Image enhancement, using the formula: l′(x, y)=log[l(x, y) / F(x, y)] to increase the image, where l(x, y) is the original image, F(x, y) is the illumination component, and l′(x, y) is the enhanced image.

[0191] Embodiment 8:

[0192] A soybean offspring identification method based on unmanned remote sensing and image recognition comprises the following steps:

[0193] S1: UAV path planning, the RGB camera is mounted on the UAV, and the UAV path is planned according to the area where the image acquisition is required;

[0194] S2: Image acquisition and preprocessing. During the flight, the RGB camera of the UAV takes low-altitude remote sensing photos of soybeans in the soybean field, obtains image data, and then preprocesses the image data;

[0195] S3: image background segmentation and optimization processing, further processing of image background segmentation and optimization;

[0196] S4: Feature extraction and recognition: extract features from the processed image, and then identify the foreign plants based on the feature extraction results.

[0197] The step S1 includes the following steps:

[0198] S11: Setting up multiple drones for collaborative operation and inputting the drone flight mission and path planning information into the control system of each drone;

[0199] S12: The Beidou positioning module built into the drone obtains the flight position of the drone in real time during flight, and controls the drone to make corrections according to the planned path if there is any deviation;

[0200] S13: The signal strength detection unit in the Beidou positioning module detects the positioning signal strength in real time;

[0201] S14: setting a lower limit threshold of the signal strength according to the positioning deviation caused by the signal strength and the required positioning deviation accuracy;

[0202] S15: When the signal strength of the positioning signal is lower than the threshold, a mutual voting-based algorithm is used to perform positioning using the networking of drones.

[0203] The step S15 comprises the following steps:

[0204] S151: Obtain theoretical positions between each UAV according to path planning;

[0205] S152: judging the actual positions of the drones according to the visual markers collected by the visual collection module;

[0206] S153: Using the deviation between the theoretical position and the actual position, voting on the position accuracy of the target UAV, and determining the reference UAV according to the voting result;

[0207] S154: The visual acquisition module acquires the visual markers on the reference UAV and calculates the orientation, and adjusts the orientation of the UAV.

[0208] The step S153 comprises the following steps:

[0209] S1531: Setting the allowable deviation coefficient k;

[0210] S1532: Decompose the orientation along the three-dimensional space to obtain the theoretical distance L of the X, Y, and Z axes respectively X , L Y , L Z And the actual distance L′ X , L′ Y , L′ Z ;

[0211] S1533: Determination ①: L′ X ∈[(1-k)L X , (1+k)L X ], ②: L′ Y ∈[(1-k)L Y , (1+k)L Y ]、③L′ X ∈[(1-k)L Z , (1+k)L Z ];

[0212] S1534: Determine the voting score based on the conditions of ①②③;

[0213] S1535: Take the average of the voting results of all drones on the target drone to obtain the voting score of the target drone.

[0214] In step S1531, k∈[0.03, 0.10], in step S1534, if ①②③ all meet the requirements, the voting result for the target drone is 1; if one of ①②③ does not meet the requirements, the voting result for the target drone is 2 / 3; if two of ①②③ do not meet the requirements, the voting result for the target drone is 1 / 3; if none of ①②③ meet the requirements, the voting result is 0.

[0215] In the step S2, the drone is a quad-rotor drone, the resolution of the RGB camera is 40 million pixels, and the flight altitude of the drone is 3.5-4.5 meters.

[0216] In the step S2, the image preprocessing includes the following steps:

[0217] S21: grayscale processing of the image, using the weighted average method for grayscale conversion, the grayscale conversion formula is Gray = 0.299*R + 0.587*G + 0.114*B, in order to reduce storage space and simplify calculation;

[0218] S22: De-noising, using extreme adaptive median filtering to remove salt and pepper noise in the image, and then using wavelet filtering to remove Gaussian noise;

[0219] S23: Image enhancement, using the formula: l′(x, y)=log[l(x, y) / F(x, y)] to increase the image, where l(x, y) is the original image, F(x, y) is the illumination component, and l′(x, y) is the enhanced image.

[0220] The step S3 comprises the following steps:

[0221] S31: using the histogram of the calculated image, calculating the intra-class variance of the foreground and background, and finding the threshold point that maximizes the inter-class variance as the segmentation point;

[0222] S32: Perform erosion and dilation operations on the segmented foreground image to remove small noise points and fill holes.

[0223] Embodiment 9:

[0224] A soybean offspring identification method based on unmanned remote sensing and image recognition comprises the following steps:

[0225] S1: UAV path planning, the RGB camera is mounted on the UAV, and the UAV path is planned according to the area where the image acquisition is required;

[0226] S2: Image acquisition and preprocessing. During the flight, the RGB camera of the UAV takes low-altitude remote sensing photos of soybeans in the soybean field, obtains image data, and then preprocesses the image data;

[0227] S3: image background segmentation and optimization processing, further processing of image background segmentation and optimization;

[0228] S4: Feature extraction and recognition: extract features from the processed image, and then identify the foreign plants based on the feature extraction results.

[0229] The step S1 includes the following steps:

[0230] S11: Setting up multiple drones for collaborative operation and inputting the drone flight mission and path planning information into the control system of each drone;

[0231] S12: The Beidou positioning module built into the drone obtains the flight position of the drone in real time during flight, and controls the drone to make corrections according to the planned path if there is any deviation;

[0232] S13: The signal strength detection unit in the Beidou positioning module detects the positioning signal strength in real time;

[0233] S14: setting a lower limit threshold of the signal strength according to the positioning deviation caused by the signal strength and the required positioning deviation accuracy;

[0234] S15: When the signal strength of the positioning signal is lower than the threshold, a mutual voting-based algorithm is used to perform positioning using the networking of drones.

[0235] The step S15 comprises the following steps:

[0236] S151: Obtain theoretical positions between each UAV according to path planning;

[0237] S152: judging the actual positions of the drones according to the visual markers collected by the visual collection module;

[0238] S153: Using the deviation between the theoretical position and the actual position, voting on the position accuracy of the target UAV, and determining the reference UAV according to the voting result;

[0239] S154: The visual acquisition module acquires the visual markers on the reference UAV and calculates the orientation, and adjusts the orientation of the UAV.

[0240] The step S153 comprises the following steps:

[0241] S1531: Setting the allowable deviation coefficient k;

[0242] S1532: Decompose the orientation along the three-dimensional space to obtain the theoretical distance L of the X, Y, and Z axes respectively X , L Y , L ZAnd the actual distance L′ X , L′ Y , L′ Z ;

[0243] S1533: Determination ①: L′ X ∈[(1-k)L X , (1+k)L X ], ②: L′ Y ∈[(1-k)L Y , (1+k)L Y ]、③L′ X ∈[(1-k)L Z , (1+k)L Z ];

[0244] S1534: Determine the voting score based on the conditions of ①②③;

[0245] S1535: Take the average of the voting results of all drones on the target drone to obtain the voting score of the target drone.

[0246] In step S1531, k∈[0.03, 0.10], in step S1534, if ①②③ all meet the requirements, the voting result for the target drone is 1; if one of ①②③ does not meet the requirements, the voting result for the target drone is 2 / 3; if two of ①②③ do not meet the requirements, the voting result for the target drone is 1 / 3; if none of ①②③ meet the requirements, the voting result is 0.

[0247] In the step S2, the drone is a quad-rotor drone, the resolution of the RGB camera is 40 million pixels, and the flight altitude of the drone is 3.5-4.5 meters.

[0248] In the step S2, the image preprocessing includes the following steps:

[0249] S21: grayscale processing of the image, using the weighted average method for grayscale conversion, the grayscale conversion formula is Gray = 0.299*R + 0.587*G + 0.114*B, in order to reduce storage space and simplify calculation;

[0250] S22: De-noising, using extreme adaptive median filtering to remove salt and pepper noise in the image, and then using wavelet filtering to remove Gaussian noise;

[0251] S23: Image enhancement, using the formula: l′(x, y)=log[l(x, y) / F(x, y)] to increase the image, where l(x, y) is the original image, F(x, y) is the illumination component, and l′(x, y) is the enhanced image.

[0252] The step S3 comprises the following steps:

[0253] S31: using the histogram of the calculated image, calculating the intra-class variance of the foreground and background, and finding the threshold point that maximizes the inter-class variance as the segmentation point;

[0254] S32: Perform erosion and dilation operations on the segmented foreground image to remove small noise points and fill holes.

[0255] The step S4 comprises the following steps:

[0256] S41: feature selection, extracting soybean leaf shape features, texture features and color features;

[0257] S42: Model construction, using lightweight convolutional neural network for feature learning and classification; the network structure includes convolution layer, depthwise separable convolution layer, batch normalization layer and ReLU activation function, as well as fully connected layer and Softmax classifier;

[0258] S43: Model training, using labeled soybean and weed image datasets for training, using cross-validation and early stopping strategies to prevent overfitting, while using learning rate decay and momentum to optimize the training process;

[0259] S44: Identification of foreign plants: input the processed image into the model, the model identifies foreign plants in the image, and outputs the identification result.

[0260] Embodiment 10:

[0261] A soybean offspring identification method based on unmanned remote sensing and image recognition comprises the following steps:

[0262] S1: UAV path planning, the RGB camera is mounted on the UAV, and the UAV path is planned according to the area where the image acquisition is required;

[0263] S2: Image acquisition and preprocessing. During the flight, the RGB camera of the UAV takes low-altitude remote sensing photos of soybeans in the soybean field, obtains image data, and then preprocesses the image data;

[0264] S3: image background segmentation and optimization processing, further processing of image background segmentation and optimization;

[0265] S4: Feature extraction and recognition: extract features from the processed image, and then identify the foreign plants based on the feature extraction results.

[0266] The step S1 includes the following steps:

[0267] S11: Setting up multiple drones for collaborative operation and inputting the drone flight mission and path planning information into the control system of each drone;

[0268] S12: The Beidou positioning module built into the drone obtains the flight position of the drone in real time during flight, and controls the drone to make corrections according to the planned path if there is any deviation;

[0269] S13: The signal strength detection unit in the Beidou positioning module detects the positioning signal strength in real time;

[0270] S14: setting a lower limit threshold of the signal strength according to the positioning deviation caused by the signal strength and the required positioning deviation accuracy;

[0271] S15: When the signal strength of the positioning signal is lower than the threshold, a mutual voting-based algorithm is used to perform positioning using the networking of drones.

[0272] The step S15 comprises the following steps:

[0273] S151: Obtain theoretical positions between each UAV according to path planning;

[0274] S152: judging the actual positions of the drones according to the visual markers collected by the visual collection module;

[0275] S153: Using the deviation between the theoretical position and the actual position, voting on the position accuracy of the target UAV, and determining the reference UAV according to the voting result;

[0276] S154: The visual acquisition module acquires the visual markers on the reference UAV and calculates the orientation, and adjusts the orientation of the UAV.

[0277] The step S153 comprises the following steps:

[0278] S1531: Setting the allowable deviation coefficient k;

[0279] S1532: Decompose the orientation along the three-dimensional space to obtain the theoretical distance L of the X, Y, and Z axes respectively X , L Y , L Z And the actual distance L′ X , L′ Y , L′ Z ;

[0280] S1533: Determination ①: L′ X ∈[(1-k)L X , (1+k)L X ], ②: L′ Y ∈[(1-k)LY , (1+k)L Y ]、③L′ X ∈[(1-k)L Z , (1+k)L Z ];

[0281] S1534: Determine the voting score based on the conditions of ①②③;

[0282] S1535: Take the average of the voting results of all drones on the target drone to obtain the voting score of the target drone.

[0283] In step S1531, k∈[0.03, 0.10], in step S1534, if ①②③ all meet the requirements, the voting result for the target drone is 1; if one of ①②③ does not meet the requirements, the voting result for the target drone is 2 / 3; if two of ①②③ do not meet the requirements, the voting result for the target drone is 1 / 3; if none of ①②③ meet the requirements, the voting result is 0.

[0284] In the step S2, the drone is a quad-rotor drone, the resolution of the RGB camera is 40 million pixels, and the flight altitude of the drone is 3.5-4.5 meters.

[0285] In the step S2, the image preprocessing includes the following steps:

[0286] S21: grayscale processing of the image, using the weighted average method for grayscale conversion, the grayscale conversion formula is Gray = 0.299*R + 0.587*G + 0.114*B, in order to reduce storage space and simplify calculation;

[0287] S22: De-noising, using extreme adaptive median filtering to remove salt and pepper noise in the image, and then using wavelet filtering to remove Gaussian noise;

[0288] S23: Image enhancement, using the formula: l′(x, y)=log[l(x, y) / F(x, y)] to increase the image, where l(x, y) is the original image, F(x, y) is the illumination component, and l′(x, y) is the enhanced image.

[0289] The step S3 comprises the following steps:

[0290] S31: using the histogram of the calculated image, calculating the intra-class variance of the foreground and background, and finding the threshold point that maximizes the inter-class variance as the segmentation point;

[0291] S32: Perform erosion and dilation operations on the segmented foreground image to remove small noise points and fill holes.

[0292] The step S4 comprises the following steps:

[0293] S41: feature selection, extracting soybean leaf shape features, texture features and color features;

[0294] S42: Model construction, using lightweight convolutional neural network for feature learning and classification; the network structure includes convolution layer, depthwise separable convolution layer, batch normalization layer and ReLU activation function, as well as fully connected layer and Softmax classifier;

[0295] S43: Model training, using labeled soybean and weed image datasets for training, using cross-validation and early stopping strategies to prevent overfitting, while using learning rate decay and momentum to optimize the training process;

[0296] S44: Identification of foreign plants: input the processed image into the model, the model identifies foreign plants in the image, and outputs the identification result.

[0297] In the step S41, the leaf shape features include area, perimeter and aspect ratio, the texture features are gray-level co-occurrence matrix or local binary value, and the color features are RGB value or HSV value.

[0298] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A soybean offspring identification method based on unmanned remote sensing and image recognition, characterized in that: The following steps are involved: S1: UAV path planning, the RGB camera is mounted on the UAV, and the UAV path is planned according to the area where the image acquisition is required; S2: Image acquisition and preprocessing. During the flight, the RGB camera of the UAV takes low-altitude remote sensing photos of soybeans in the soybean field, obtains image data, and then preprocesses the image data; S3: image background segmentation and optimization processing, further processing of image background segmentation and optimization; S4: Feature extraction and recognition: extract features from the processed image, and then identify the foreign plants based on the feature extraction results.

2. The soybean offal identification method based on unmanned remote sensing and image recognition according to claim 1, characterized in that: The step S1 includes the following steps: S11: Setting up multiple drones for collaborative operation and inputting the drone flight mission and path planning information into the control system of each drone; S12: The Beidou positioning module built into the drone obtains the flight position of the drone in real time during flight, and controls the drone to make corrections according to the planned path if there is any deviation; S13: The signal strength detection unit in the Beidou positioning module detects the positioning signal strength in real time; S14: setting a lower limit threshold of the signal strength according to the positioning deviation caused by the signal strength and the required positioning deviation accuracy; S15: When the signal strength of the positioning signal is lower than the threshold, a mutual voting-based algorithm is used to perform positioning using the networking of drones.

3. The soybean offal identification method based on unmanned remote sensing and image recognition according to claim 2, characterized in that: The step S15 comprises the following steps: S151: Obtain theoretical positions between each UAV according to path planning; S152: judging the actual positions of the drones according to the visual markers collected by the visual collection module; S153: Using the deviation between the theoretical position and the actual position, voting on the position accuracy of the target UAV, and determining the reference UAV according to the voting result; S154: The visual acquisition module acquires the visual markers on the reference UAV and calculates the orientation, and adjusts the orientation of the UAV.

4. The soybean offal identification method based on unmanned remote sensing and image recognition according to claim 3, characterized in that: The step S153 comprises the following steps: S1531: Setting the allowable deviation coefficient k; S1532: Decompose the orientation along the three-dimensional space to obtain the theoretical distance L of the X, Y, and Z axes respectively X , L Y , L Z And the actual distance L′ X , L′ Y , L′ Z ; S1533: Determination ①: L′ X ∈[(1-k)L X , (1+k)L X ], ②: L′ Y ∈[(1-k)L Y , (1+k)L Y ]、③L′ X ∈[(1-k)L Z , (1+k)L Z ]; S1534: Determine the voting score based on the conditions of ①②③; S1535: Take the average of the voting results of all drones on the target drone to obtain the voting score of the target drone.

5. The soybean offal identification method based on unmanned remote sensing and image recognition according to claim 4, characterized in that: In step S1531, k∈[0.03, 0.10], in step S1534, if ①②③ all meet the requirements, the voting result for the target drone is 1; if one of ①②③ does not meet the requirements, the voting result for the target drone is 2 / 3; if two of ①②③ do not meet the requirements, the voting result for the target drone is 1 / 3; if none of ①②③ meet the requirements, the voting result is 0.

6. The soybean offal identification method based on unmanned remote sensing and image recognition according to claim 1, characterized in that: In the step S2, the drone is a quad-rotor drone, the resolution of the RGB camera is 40 million pixels, and the flight altitude of the drone is 3.5-4.5 meters.

7. The soybean offal identification method based on unmanned remote sensing and image recognition according to claim 1, characterized in that: In step S2, the image preprocessing includes the following steps: S21: grayscale processing of the image, using the weighted average method for grayscale conversion, the grayscale conversion formula is Gray = 0.299*R + 0.587*G + 0.114*B, in order to reduce storage space and simplify calculation; S22: De-noising, using extreme adaptive median filtering to remove salt and pepper noise in the image, and then using wavelet filtering to remove Gaussian noise; S23: Image enhancement, using the formula: l′(x, y)=log[l(x, y) / F(x, y)] to increase the image, where l(x, y) is the original image, F(x, y) is the illumination component, and l′(x, y) is the enhanced image.

8. The soybean offal identification method based on unmanned remote sensing and image recognition according to claim 1, characterized in that: The step S3 comprises the following steps: S31: using the histogram of the calculated image, calculating the intra-class variance of the foreground and background, and finding the threshold point that maximizes the inter-class variance as the segmentation point; S32: Perform erosion and dilation operations on the segmented foreground image to remove small noise points and fill holes.

9. The soybean offal identification method based on unmanned remote sensing and image recognition according to claim 1, characterized in that: The step S4 comprises the following steps: S41: feature selection, extracting soybean leaf shape features, texture features and color features; S42: Model construction, using lightweight convolutional neural network for feature learning and classification; the network structure includes convolution layer, depthwise separable convolution layer, batch normalization layer and ReLU activation function, as well as fully connected layer and Softmax classifier; S43: Model training, using labeled soybean and weed image datasets for training, using cross-validation and early stopping strategies to prevent overfitting, while using learning rate decay and momentum to optimize the training process; S44: Identification of foreign plants: input the processed image into the model, the model identifies foreign plants in the image, and outputs the identification result.

10. The soybean offal identification method based on unmanned remote sensing and image recognition according to claim 9, characterized in that: In the step S41, the leaf shape features include area, perimeter and aspect ratio, the texture features are gray-level co-occurrence matrix or local binary value, and the color features are RGB value or HSV value.

Images