CN116721060A - Method for detecting red palm weevil beetles based on attention mechanism - Google Patents
Method for detecting red palm weevil beetles based on attention mechanism Download PDFInfo
- Publication number
- CN116721060A CN116721060A CN202310555293.9A CN202310555293A CN116721060A CN 116721060 A CN116721060 A CN 116721060A CN 202310555293 A CN202310555293 A CN 202310555293A CN 116721060 A CN116721060 A CN 116721060A
- Authority
- CN
- China
- Prior art keywords
- network
- target detection
- red palm
- training
- detection network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 241001078693 Rhynchophorus ferrugineus Species 0.000 title claims abstract description 39
- 230000007246 mechanism Effects 0.000 title claims abstract description 37
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000001514 detection method Methods 0.000 claims abstract description 126
- 238000012549 training Methods 0.000 claims abstract description 43
- 241001677259 Acanthophoenix rubra Species 0.000 claims abstract description 35
- 241000254173 Coleoptera Species 0.000 claims abstract description 35
- 238000012360 testing method Methods 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 12
- 238000005516 engineering process Methods 0.000 claims description 6
- 238000013135 deep learning Methods 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 claims description 2
- 230000005284 excitation Effects 0.000 claims description 2
- 238000001125 extrusion Methods 0.000 claims description 2
- 230000001629 suppression Effects 0.000 claims description 2
- 238000002474 experimental method Methods 0.000 claims 1
- 241000607479 Yersinia pestis Species 0.000 description 11
- 241000196324 Embryophyta Species 0.000 description 7
- 230000002265 prevention Effects 0.000 description 7
- 210000003739 neck Anatomy 0.000 description 5
- 244000060011 Cocos nucifera Species 0.000 description 4
- 235000013162 Cocos nucifera Nutrition 0.000 description 4
- 241000238631 Hexapoda Species 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000002776 aggregation Effects 0.000 description 3
- 238000004220 aggregation Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000012546 transfer Methods 0.000 description 3
- 230000006378 damage Effects 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 244000080767 Areca catechu Species 0.000 description 1
- 235000006226 Areca catechu Nutrition 0.000 description 1
- 208000032544 Cicatrix Diseases 0.000 description 1
- 240000003133 Elaeis guineensis Species 0.000 description 1
- 235000001950 Elaeis guineensis Nutrition 0.000 description 1
- 244000104275 Phoenix dactylifera Species 0.000 description 1
- 235000010659 Phoenix dactylifera Nutrition 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000002555 auscultation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001066 destructive effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 238000005304 joining Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000008654 plant damage Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 231100000241 scar Toxicity 0.000 description 1
- 230000037387 scars Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 241000894007 species Species 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Abstract
The invention provides a method for detecting red palm weevil beetles based on an attention mechanism, which comprises the following steps: constructing a target detection data set, namely a red palm beetle data set, and carrying out data enhancement and expansion on the target detection data set; constructing a target detection network, and training the constructed target detection network by utilizing a data set; the target detection network can extract and learn the characteristics of the red palm beetles and finish the output detection of the red palm beetle individuals; the trained target detection network can be used for detecting and identifying the red palm weevil beetles in different scenes; improving the target detection network, and adding an attention mechanism into a backbone network of the target detection network to obtain an improved target detection network for the red palm beetles; retraining the data set of the red palm weevil, and comparing the accuracy, precision and the like of the identification of the red palm weevil under the original target detection network and the improved target detection network. By adopting the method, the accuracy of target detection of the red palm weevil can be remarkably improved.
Description
Technical Field
The invention relates to the technical field of target detection of machine learning, in particular to a method for detecting red palm weevil beetles based on an attention mechanism.
Background
The red palm weevil is a destructive pest of coconut trees and also damages palm plants such as oil palm, betel nut, date palm and the like. It is distributed in Indonesia, malaysia, philippines, thailand, burmese and China. The insect is found to be harmful in the provinces of Hainan, guangdong, taiwan, yunnan and the like in China. The adult red palm weevil is generally not directly harmful and mainly takes the stem and the crown and the leaf of the tree as the insect of the larvae. Early adults spawn and hatch in scars, cracks and fissures near the crowns of plants, and the larvae are bored into tender tissues of the top stems of the pests, once victimized, the damage is serious. Early harmful is hard to be perceived, and later harmful is easy to be seen. When the pest is at first, the blade of the new suction is incomplete, the ear or the medical stethoscope is used for being close to the stem of the damaged tree, the eating sound of the sand in the stem of the larva can be heard, and in the later period of the pest, the central blade is dried up, and waste fiber scraps or brown viscous liquid are discharged from the decayed holes. The serious plant is damaged, the new She Diaowei plant is dead, and only a plurality of old leaves are left, so that the plant is difficult to rescue. Some trunks are even eaten hollow, leaving only empty shells. Once harmful, it can lead to death of the coconut forest or palmaceae plant pieces, and there is a constant trend to spread. Once the pest is harmful to the plantation, if no control measures are taken, the pest will continue to expand and spread until the entire plantation is destroyed.
Therefore, the pest control should adopt the comprehensive control measures mainly including prevention, firstly, depending on the artificial observation of whether the crown of the plant has abnormal variation, the stem of the plant is inspected and listened by a common stethoscope, and once the fibrous scraps are found at the heart leaves of the crown or the eating sound of the trunk is heard, the pest is immediately controlled. After manual observation and auscultation, the prevention measures adopted are strictly preventing plant injury, removing or reducing insect sources in the garden, chemical prevention, trapping and killing and the like. However, the cost of the manual prevention and control mode in the early stage is high and the efficiency is low, so that early discovery of red palm weevil pests becomes a great research hot spot and difficulty in order to ensure that palm plants are protected efficiently.
In some cities in the south of China, coconuts and other palm plants are ideal environment greening tree species, the project of developing palm plants and million mu coconut forests in regions such as Yunnan of China is strong, the defects of the traditional red palm beetle prevention and control technology are overcome when the palm plants are required to be protected from insect pests effectively, and an accurate and efficient red palm beetle detection technology is searched out by combining with the artificial intelligence emerging technology so as to achieve the aim of overall pest prevention and control.
Disclosure of Invention
The embodiment of the invention provides a detection method for red palm weevil based on an attention mechanism, which can solve the problems of high cost and low efficiency of the manual prevention and control method for red palm weevil in the prior art. The technical scheme is as follows:
the embodiment of the invention provides a method for detecting red palm weevil beetles based on an attention mechanism, which comprises the following steps:
collecting an image of red palm weevil, performing processing such as enhancement marking on the image, and manufacturing a data set;
constructing a target detection network, and training the constructed target detection network by utilizing the data set; the constructed target detection network adopts a YOLOv5 deep learning framework;
performing algorithm improvement on the original YOLOv5 network, adding an attention mechanism into the backbone network of the original YOLOv5 network, and retraining the data set; among them, the attention mechanism is a Squeeze-and-incentive network (Squeeze-and-ExcitationNet, SENet), which is the most common one, belonging to channel attention; the method strengthens the characteristics of the input image, so that the model pays more attention to the typical characteristics with large information quantity and ignores the secondary characteristics with small information quantity; the method mainly comprises extrusion and excitation, wherein the method comprises the steps of firstly compressing the characteristics of an input image, then carrying out characteristic learning on a compressed characteristic image to obtain learning weights, and finally multiplying the learning weights by an original characteristic image to obtain final characteristics;
further, the dataset comprises: a training set and a testing set;
the acquiring the image of the red palm weevil and the making of the data set comprises the following steps:
at different moments of the day, the camera device is used for collecting images of red palm beetles in different states in the field environment, the red palm beetles in the images are marked by using the marking tool and recorded in the marking file, a part of the images and the corresponding marking files are randomly selected to serve as a training set, and the rest of the images and the corresponding marking files serve as a test set.
Further, the constructing the target detection network, and the training the constructed target detection network by using the data set includes:
selecting a YOLOv5 target detection algorithm, and constructing a target detection network, wherein the target detection network comprises: the system comprises an input end, a backbone network, a neck network and a detection head (output end), wherein the structure of the characteristic network is based on a selected YOLOv5 network;
and inputting the data set, training the detection network by using a deep learning technology, combining the feature network and the loss function of the detection network, and performing fine tuning training on the whole target detection network by using the data set.
Further, joining the attention mechanism, utilizing the improved detection network technique includes:
adding a squeezing and exciting network (Squeeze-and-ExcitationNet, SENet), namely SE attention mechanism, into a target detection network YOLOv5 model, wherein after the backbone network is improved, the attention mechanism is added to obtain an improved target detection model,
using generalized cross joint loss (GIOULoss) as a loss function of frame regression (Boundingbox), and simultaneously requiring non-maximum suppression (NonMaximumSuppression, NMS) operation to complete output detection;
and (3) after calculating the loss function value of the feature network, performing back propagation, wherein the test is performed on the test set after each training round, if the loss function value of the feature network on the test set is increased, the training is finished in advance, otherwise, the training is continued until the current iteration number is greater than or equal to the preset training number.
Further, retraining and testing the improved model to obtain indexes such as detection accuracy, recall rate and the like;
further, the front and rear model detection performance was evaluated:
evaluation indexes of the detection Precision include a cross ratio (IntersectionofUnion, ioU), an accuracy (P), a Recall (R), and an average accuracy mean value (meanAveragePrecision, mAP), which are expressed as:
where TP represents the number of positive samples detected as positive samples; FP represents the number of negative samples detected as positive samples; FN represents the number of positive samples detected as negative samples; n is the number of detection sample categories. Ba represents the region of the predicted frame, bb represents the region of the real frame; the overlap ratio IoU represents the overlapping degree of the real frame and the predicted frame, and the higher the value is, the higher the overlapping degree is, and the more accurate the predicted result is; when the IoU threshold value is 0.5, the mAP at the moment is marked as mAP@.5; and (5) starting the IoU threshold value from 0.5 to 0.95, adding 0.05 each time, taking 10 threshold values, respectively calculating corresponding average precision mean values, and then calculating the average value, wherein the average value is recorded as mAP@5:95.
Further, the identification accuracy of the original target detection network and the improved network is compared, and a result is obtained.
Further, the method further comprises:
in the improved network model, the data set of the red palm beetles is trained, indexes are comprehensively evaluated, and the accuracy, precision and the like of the identification of the red palm beetles under the original target detection network and the improved target detection network are compared.
The technical scheme provided by the embodiment of the invention has the beneficial effects that at least:
the invention carries out feature extraction and training on the image of the red palm beetle through the target detection network, improves the target detection network, continues training after adding the attention mechanism, tests the model after training, and compares the results. Compared with the prior art, the method has the advantages that the detection result of the red palm weevil is more accurate, the detection is more convenient and faster, and the method has better robustness for the detection in a field shielding scene, so that the problems of high cost and low efficiency of manually preventing and controlling the red palm weevil in the prior art are solved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of a method for detecting red palm weevil based on an attention mechanism according to an embodiment of the present invention;
FIG. 2 is a detailed flow chart of a method for detecting red palm weevil based on an attention mechanism according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a YOLOv5 target detection network according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of the SE attention mechanism according to an embodiment of the present invention;
fig. 5 is a schematic diagram of a spectrum trap model to be built according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Examples:
as shown in fig. 1-5, an embodiment of the present invention provides a method for detecting beetles of red palm image based on an attention mechanism, the method comprising:
s101, acquiring an image of red palm weevil and making a data set;
s102, constructing a target detection network, and training the constructed target detection network by utilizing the data set; the target detection network can extract and learn the characteristics of the red palm beetles and finish the output detection of the red palm beetle individuals;
s103, detecting and identifying the red palm weevil beetles in different scenes by using a trained target detection network; improving the target detection network, and adding an attention mechanism into a backbone network of the target detection network to obtain an improved target detection network for the red palm beetles; retraining and detecting the data set of the red palm weevil, and comparing the accuracy and precision of the identification of the red palm weevil under the original target detection network and the improved target detection network.
According to the method for detecting the red palm beetles based on the attention mechanism, disclosed by the embodiment of the invention, the characteristics of the images of the red palm beetles are extracted and trained through the target detection network, the target detection network is improved, the training is continued after the attention mechanism is added, and the test and the comparison result are carried out on the trained model. Compared with the prior art, the method has the advantages that the detection result of the red palm weevil is more accurate, the detection is more convenient and faster, and the method has better robustness for the detection in a field shielding scene, so that the problems of high cost and low efficiency of manually preventing and controlling the red palm weevil in the prior art are solved.
In this embodiment, the target detection network is a YOLOv5 target detection network in a one-stage target detection network.
In a specific embodiment of the foregoing method for detecting beetles in red palm based on an attention mechanism, further, the data set includes: a training set and a testing set;
the acquiring the image of the red palm weevil and the making of the data set comprises the following steps:
at different moments of the day, the camera device is used for collecting images of red palm beetles in different states in the field environment, the red palm beetles in the images are marked by using the marking tool and recorded in the marking file, a part of the images and the corresponding marking files are randomly selected to serve as a training set, and the rest of the images and the corresponding marking files serve as a test set.
In this embodiment, for example, a camera may be used to capture an image of the red palm beetle, then an image of the red palm beetle in a field environment is collected, and a labeling tool labelimg is used to label the red palm beetle in the image, label data is saved as an XML file (i.e., a label), 90% of the image and the corresponding label file are randomly selected as a training set, and the remaining image and the corresponding label file are used as a test set.
In the specific embodiment of the foregoing method for detecting red palm weevil based on the attention mechanism, further, as shown in fig. 2, 3 and 4, the constructing the target detection network, training the constructed target detection network using the data set includes:
selecting a YOLOv5 target detection algorithm, and constructing a target detection network, wherein the target detection network comprises: the device comprises an input end, a Backbone network of backbones, a Neck network of Necks and a Head detection Head (output end), wherein the structure of the characteristic network is based on a selected YOLOv5 network;
and inputting the data set, training the detection network by using a deep learning technology, combining the feature network and the loss function of the detection network, and performing fine tuning training on the whole target detection network by using the data set.
In the embodiment, a feature network is firstly constructed, the input of the feature network is an original image, and the output is a feature vector of the image; then, a detection network is constructed on the basis of the characteristic network, the input of the detection network is two characteristic vectors before and after a characteristic network pooling layer, and the output is a target detection result, namely: target individuals of the red palm weevil beetles.
In this example, the specific flow of detection is shown in fig. 2, firstly, a data set is made, the target detection network is trained and then input into the target detection system, secondly, the obtained red palm beetle data are transmitted into the YOLOv5 feature extraction network, feature extraction, up-sampling and feature fusion are performed on the input data in the network, prediction information is generated through regression analysis and the like, and the prediction information is transmitted to the output end to output.
In this embodiment, the selected target detection network is the target detection network YOLOv5 at one stage.
In this embodiment, in order to better understand YOLOv5, a target detection algorithm is described first; common target detection algorithms can be divided into two categories: the method is a two-stage target detection algorithm represented by RCNN, such as R-CNN, fastR-CNN, fasterR-CNN and the like, wherein candidate frames are generated by the algorithm firstly, a convolution network is executed again to identify a detection object, and the algorithm has high detection precision but slower speed; the other type is a one-stage target detection algorithm represented by SSD and YOLO series, the algorithm directly generates class probability and position information of an object, a final detection result can be directly obtained through single detection, and compared with a two-stage algorithm, the algorithm has higher speed but has lost precision; in the one-stage detection algorithm, the running speed of the SSD neural network is lower than that of YOLO, the detection precision is slightly lower than that of Faster-CNN, and the stability is poor; in the YOLO series, YOLOv5 is one of the advanced algorithms at present, has the advantages of high average detection speed, high flexibility, quick deployment and the like, is widely applied, and can be suitable for quick real-time detection of red palm weevil in a natural complex environment.
In this embodiment, YOLOv5 belongs to a one-stage target detection algorithm, and has the characteristics of flexible and changeable model scale, low deployment cost, faster training time and reasoning speed, and the like; YOLOv5 is a single-stage target detection algorithm, and a new improvement thought is added to the algorithm on the basis of YOLOv4, so that the speed and the precision of the algorithm are greatly improved; the YOLOv5 network model mainly comprises an input end, a Backbone network of a backhaul, a Neck network and a Head output end, and a network schematic diagram of the YOLOv5 network model is shown in figure 3; an input end: the part comprises an image preprocessing stage, wherein an input image is scaled to the input size of a network, normalization and other operations are performed, and the operation method comprises a Mosaic data enhancement operation, a self-adaptive anchor frame calculation and a self-adaptive image scaling method; the mosaics data enhancement sequentially combines the four pictures to achieve the effect of enriching the picture background; the self-adaptive picture scaling is to add black edges to the original picture in a self-adaptive way, uniformly scale the original picture to a standard size and send the standard size into a detection network; the self-adaptive anchor frame is to calculate the best anchor frame value in the training set in a self-adaptive way when each training is performed; YOLOv5 adopts CSPDarknet53 as backbone network, and its main function is to extract characteristic information in the picture for subsequent network use.
The Neck network is composed of an SPPF module, a feature pyramid network and a path aggregation network and is positioned in the middle of the backbone network and the output end, so that the diversity and the robustness of the features can be further improved; the structure mainly solves the problem of multi-scale detection in target detection, the feature pyramid network carries out feature transfer fusion in three modes of using a top-down mode and a top-down mode to transversely connect high-level feature information, and improves the feature information capacity of the final network layer, but the feature pyramid network only enhances semantic information transfer and is weaker in shallow positioning information transfer; the path aggregation network is characterized in that a pyramid from bottom to top is added after the characteristic pyramid network is convolved by 3 multiplied by 3, so that the transmission of positioning information is enhanced; YOLOv5 takes as output 76 x 76 features generated by the feature pyramid; meanwhile, the characteristics of 38 multiplied by 38 and 19 multiplied by 19 generated by the path aggregation network are taken as output, and the anchor frame is combined to finish the detection of targets with different scales; the generalized cross combination is adopted as a loss function of frame regression at the output end, and in the post-processing process of target detection, non-maximum value inhibition operation is usually required for screening a plurality of target frames, and the generated three feature maps with different sizes are used for calculating the following loss function.
In this embodiment, the added attention mechanism is shown in fig. 4, and is an attention mechanism of a channel type, that is, the attention mechanism is added in the channel dimension, and the main contents are squeeze and specification.
On the basis of an original learning mechanism, a new network path is opened up, the attention degree of each channel in the feature map is obtained through operation, and an attention weight is configured for each feature channel according to the attention degree, so that the convolution network focuses on the feature channels more, the channels of the feature map which are useful for the current task are realized, and the feature channels which are less useful for the current task are restrained; mainly comprises two operations: extruding and exciting; the SE module first performs a squeeze operation on the feature map obtained by convolution to obtain channel-level global features, and then performs a stimulus operation on the global features to obtain weights for different channels and relationships between the channels.
The invention and its embodiments have been described above with no limitation, and the actual construction is not limited to the embodiments of the invention as shown in the drawings. In summary, if one of ordinary skill in the art is informed by this disclosure, a structural manner and an embodiment similar to the technical solution should not be creatively devised without departing from the gist of the present invention.
Claims (7)
1. The method for detecting the red palm weevil beetles based on the attention mechanism is characterized by comprising the following steps of:
collecting images of red palm weevil and preparing a data set;
constructing a target detection network, and training the constructed target detection network by utilizing the data set; the target detection network can extract and learn the characteristics of the red palm beetles and finish the output detection of the red palm beetle individuals;
the trained target detection network can be used for detecting and identifying the red palm weevil beetles in different scenes; improving the target detection network, and adding an attention mechanism into a backbone network of the target detection network to obtain an improved target detection network for the red palm beetles; retraining the data set of the red palm weevil, and comparing the accuracy and precision of the identification of the red palm weevil under the original target detection network and the improved target detection network.
2. The method for detecting beetles in red palm based on an attention mechanism according to claim 1, wherein said data set comprises: a training set and a testing set;
the image of the red palm weevil, the data set is made comprising:
at different moments of the day, the camera device is used for collecting images of red palm beetles in different states in the field environment, the red palm beetles in the images are marked by using the marking tool and recorded in the marking file, a part of the images and the corresponding marking files are randomly selected to serve as a training set, and the rest of the images and the corresponding marking files serve as a test set.
3. The method for detecting beetles in red palm based on an attention mechanism according to claim 2, wherein said constructing a target detection network, training the constructed target detection network using the data set comprises:
selecting a YOLOv5 target detection algorithm, and constructing a target detection network, wherein the target detection network comprises: the device comprises an input end, a backbone network, a neck network and a detection head, wherein the structure of the characteristic network is based on a selected YOLOv5 network;
and inputting the data set, training the detection network by using a deep learning technology, combining the feature network and the loss function of the detection network, and performing fine tuning training on the whole target detection network by using the data set.
4. The method for detecting beetles in red palm based on an attention mechanism according to claim 3, wherein adding to the attention mechanism, utilizing an improved detection network technique comprises:
adding an extrusion and excitation network, namely SE attention mechanism, into a target detection network YOLOv5 model, wherein after the backbone network is improved, adding the attention mechanism to obtain an improved target detection model,
the generalized cross joint loss is used as a loss function of frame regression, and meanwhile, non-maximum suppression operation is required to complete output detection;
and (3) after calculating the loss function value of the feature network, performing back propagation, wherein the test is performed on the test set after each training round, if the loss function value of the feature network on the test set is increased, the training is finished in advance, otherwise, the training is continued until the current iteration number is greater than or equal to the preset training number.
5. The method for detecting the red palm weevil based on the attention mechanism according to claim 4, wherein the evaluation indexes of the detection precision are a cross ratio IoU, an accuracy rate P, a recall rate R and an average accuracy mean mAP, which are expressed as:
where TP represents the number of positive samples detected as positive samples; FP represents the number of negative samples detected as positive samples; FN represents the number of positive samples detected as negative samples; n is the number of detection sample categories; ba represents the region of the predicted frame, bb represents the region of the real frame; the overlap ratio IoU represents the overlapping degree of the real frame and the predicted frame, and the higher the value is, the higher the overlapping degree is, and the more accurate the predicted result is; when the IoU threshold value is 0.5, the mAP at the moment is marked as mAP@.5; and (5) starting the IoU threshold value from 0.5 to 0.95, adding 0.05 each time, taking 10 threshold values, respectively calculating corresponding average precision mean values, and then calculating the average value, wherein the average value is recorded as mAP@5:95.
6. The method for detecting beetles in red palm based on an attention mechanism of claim 4, wherein said training the data set of beetles in red palm based on the improved feature detection network comprises:
the weight parameters of a network training model are fixed, the front training and the back training are both based on a Pytorch deep learning framework, an operating system adopted by the model training is Windows10, a hardware platform is an NVIDIAGeforceGTX1070 display card, and a memory 8G is operated; the parallel computing framework with the version number of CUDA10.2 is used;
fixing training times, and performing experiment total training times for 1000 times; model training parameters were set as follows: learning rate=0.01, momentum=0.937, decay=0.0005, lot=8.
7. The method for detecting beetles in red palm based on an attention mechanism according to claim 5, wherein said method further comprises:
in the improved network model, the data set of the red palm weevil is trained and tested, the index is comprehensively evaluated, and the accuracy, precision and the like of the identification of the red palm weevil under the original target detection network and the improved target detection network are compared.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310555293.9A CN116721060A (en) | 2023-08-08 | 2023-08-08 | Method for detecting red palm weevil beetles based on attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310555293.9A CN116721060A (en) | 2023-08-08 | 2023-08-08 | Method for detecting red palm weevil beetles based on attention mechanism |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116721060A true CN116721060A (en) | 2023-09-08 |
Family
ID=87866940
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310555293.9A Pending CN116721060A (en) | 2023-08-08 | 2023-08-08 | Method for detecting red palm weevil beetles based on attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116721060A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117197148A (en) * | 2023-11-08 | 2023-12-08 | 天津海关动植物与食品检测中心 | Thermal imaging-based wood boring pest detection method, system and medium |
-
2023
- 2023-08-08 CN CN202310555293.9A patent/CN116721060A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117197148A (en) * | 2023-11-08 | 2023-12-08 | 天津海关动植物与食品检测中心 | Thermal imaging-based wood boring pest detection method, system and medium |
| CN117197148B (en) * | 2023-11-08 | 2024-02-02 | 天津海关动植物与食品检测中心 | Thermal imaging-based wood boring pest detection method, system and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Khan et al. | Deep diagnosis: A real-time apple leaf disease detection system based on deep learning | |
| CN107016405B (en) | A kind of pest image classification method based on classification prediction convolutional neural networks | |
| Khan et al. | Deep learning for apple diseases: classification and identification | |
| CN114445785B (en) | Internet of things-based litchi insect pest monitoring and early warning method and system and storage medium | |
| CN110717903A (en) | Method for detecting crop diseases by using computer vision technology | |
| CN110245720B (en) | Deep learning-based intelligent citrus pest diagnosis method and system | |
| CN107622233A (en) | A kind of Table recognition method, identifying system and computer installation | |
| CN114708231A (en) | Sugarcane aphid target detection method based on light-weight YOLO v5 | |
| CN116721060A (en) | Method for detecting red palm weevil beetles based on attention mechanism | |
| CN114972208B (en) | YOLOv 4-based lightweight wheat scab detection method | |
| CN116471930A (en) | Intelligent mosquito catcher for classifying mosquitoes | |
| CN112966621A (en) | Citrus insect damage identification method based on deep convolutional network and transfer learning | |
| CN109614896A (en) | A method of the video content semantic understanding based on recursive convolution neural network | |
| Al-Saqer et al. | Artificial neural networks based red palm weevil (Rynchophorus Ferrugineous, Olivier) recognition system | |
| Chulu et al. | A convolutional neural network for automatic identification and classification of fall army worm moth | |
| Sehree et al. | Olive trees cases classification based on deep convolutional neural network from unmanned aerial vehicle imagery | |
| Ibrahim et al. | Palm leaf nutrient deficiency detection using convolutional neural network (CNN) | |
| Goyal et al. | Plant species identification using leaf image retrieval: A study | |
| Chiu et al. | Semantic segmentation of lotus leaves in UAV aerial images via U-Net and deepLab-based networks | |
| Brar et al. | A smart approach to coconut leaf spot disease classification using computer vision and deep learning technique | |
| Ibarra et al. | Detection of Panama Disease on Banana Leaves Using the YOLOv4 Algorithm | |
| Watcharabutsarakham et al. | An approach for density monitoring of brown planthopper population in simulated paddy fields | |
| Liang | Effectiveness of transfer learning, convolutional neural network and standard machine learning in computer vision assisted bee health assessment | |
| Gunes et al. | Classification of hazelnuts with CNN based deep learning system | |
| Choi et al. | Classification of growth conditions in paprika leaf using deep neural network and hyperspectral images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |