CN-121982547-A - Pepper pest detection method based on improvement YOLOv (automatic control unit)

CN121982547ACN 121982547 ACN121982547 ACN 121982547ACN-121982547-A

Abstract

The invention discloses a pepper insect pest detection method based on an improvement YOLOv <8 >, which comprises the following steps of S1, collecting pepper insect pest images, S2, processing the collected pepper insect pest images, dividing processed image data into a training set, a verification set and a test set, constructing a standardized pepper insect pest detection data set, S3, inputting the pepper insect pest detection data set into an improved YOLOv detection network, and outputting a pepper insect pest detection result after processing through the improved OLOv n detection network, wherein the detection result comprises insect pest category information, corresponding confidence and target boundary box positions.

Inventors

DAI MIN
MIAO HONG
HU CHIYU
YUAN HUALI
SONG TAO
Zong Zhiyang
Tang futian
WANG ZIXU
ZHENG JIEWEN
Cheng Jiawang

Assignees

扬州大学

Dates

Publication Date: 20260505
Application Date: 20260206

Claims (6)

1. The pepper insect pest detection method based on the improvement YOLOv is characterized by comprising the following steps of, S1, collecting a pepper insect pest image; S2, processing the collected pepper insect pest images, and dividing the processed image data into a training set, a verification set and a test set; S3, inputting the pepper insect pest detection data set into an improved YOLOv detection network, and outputting a pepper insect pest detection result after processing by the improved YOLOv n detection network, wherein the detection result comprises insect pest type information, corresponding confidence and target boundary box positions.
2. The method for detecting pepper insect pests based on the improvement YOLOv as claimed in claim 1, wherein in S2, the pretreatment step is, Performing optical transformation and geometric transformation on the collected pepper insect pest image to obtain a preprocessed image; Marking the preprocessed image by using a marking tool to generate an image containing insect pest type information and marking data corresponding to the target position; And dividing the image data subjected to pretreatment and labeling into a training set, a verification set and a test set according to a preset proportion.
3. The pepper insect pest detection method based on the improvement YOLOv as claimed in claim 1, wherein step S3 is specifically, S301, carrying out feature extraction on an input image through a backbone network to generate a multi-scale backbone feature map; S302, inputting a trunk feature map into a neck network, introducing a cross-scale context fusion module CCFM into the neck network, and carrying out joint modeling on different scale features to obtain a fusion feature map; S303, inputting the fusion feature map into a detection head module to generate detection features for target classification and bounding box regression; S304, calculating a regression error between a prediction frame and a real frame by adopting an Inner-MPDIOU boundary frame regression loss function based on the prediction result output in the step S303 and combining the real label information corresponding to the training set image, constructing a total loss function by combining the classification loss and the confidence loss, updating model parameters by a back propagation mode, and finishing training of improving the YOLOv n detection model by loop iteration to obtain a final model weight; S305, loading the model weight trained in the step S304 into a modified YOLOv detection network, and inputting a testing set or a pepper insect pest image to be detected into the modified YOLOv detection network.
4. The pepper insect pest detection method based on improvement YOLOv as set forth in claim 3, wherein in step S301, a CSPHet structure is adopted in the backbone network, and a heterogeneous convolution kernel combination and cross-dimension feature interaction mechanism is introduced through the CSPHet structure, so that the backbone network can extract fine-grained features of small objects of pepper insect pests more effectively under a complex background condition.
5. The pepper insect pest detection method based on improvement YOLOv as claimed in claim 3, wherein DyHead dynamic detection head structure is adopted in the detection head module, and the dynamic weighting mechanism is as follows, ; Wherein F l is DyHead The fusion feature map of the layer DConv k is a deformable convolution operation, alpha k is a weight coefficient of the kth layer feature obtained by a DyHead dynamic weighting mechanism according to input feature self-adaptive learning when being fused to the first layer output feature, l is a current target level, and X k is the kth layer output feature An input feature map of a layer; DyHead a cross-level context enhancement module is introduced into the dynamic detection head, and the formula of the dynamic activation function in DyHead is that, ; Where Y is the output profile, a 1 (X) 、a 2 (X) is the input dependent scaling factor, and b 1 (X)、b 2 (X) is the input dependent bias.
6. The method for detecting pepper insect damage based on improvement YOLOv as recited in claim 3, wherein in step S304, the I nner-MPDIOU bounding box regression loss function is, ; Wherein IoU is the intersection ratio between the predicted frame and the real frame, d c inner is the internal center distance, d b inner is the multi-view internal boundary distance, d s inner is the shape consistency term, and λ c 、λ b and λ s are the weight coefficients of the constraint terms respectively.

Description

Pepper pest detection method based on improvement YOLOv (automatic control unit) Technical Field The invention relates to the technical field of pest and disease damage identification, in particular to a pepper pest and disease damage detection method based on improvement YOLOv. Background The capsicum is taken as one of important cash crops in China and worldwide, and is easy to attack by various insect pests, such as aphids, thrips, cabbage caterpillars and the like in the planting process. The occurrence of insect damage not only can cause the growth resistance and the yield reduction of pepper plants, but also can seriously influence the quality of fruits, thereby bringing great economic loss to agricultural production. Therefore, the method for identifying and monitoring the pepper insect pests rapidly and accurately is an important precondition for realizing scientific prevention and control and accurate medication. The traditional pepper insect pest identification mode mainly depends on manual field inspection or experience judgment, and the method has the advantages of high labor intensity, low efficiency and large influence on subjective experience, and is difficult to meet the large-scale and real-time insect pest monitoring requirements. With the development of computer vision and artificial intelligence technology, pest identification methods based on image processing and deep learning are attracting attention. The target detection algorithm can locate and identify the insect pest targets under a complex background, and provides a new technical means for intelligent monitoring of agricultural insect diseases. In the existing research, YOLO (You Only Look Once) series target detection algorithm is widely applied to the field of crop pest identification due to the advantages of high detection speed, end-to-end training and the like. However, in an actual pepper planting scene, the pest targets generally have the characteristics of small volume, dense distribution, similar shape, easy influence of illumination and shielding, and the like, so that the existing YOLO algorithm still has defects in the aspects of small target pest detection precision and robustness. For example, in the prior art, a corn leaf disease detection system and method based on an improved YOLO algorithm are disclosed, publication number is CN120894775A, publication date is 2025, invention patent of 11 month 04, the disease detection method improves detection precision and speed, reduces model calculation cost, improves system deployment and practical capability, and is suitable for an automatic disease detection scene of an agricultural field. But the method does not effectively solve the problems of insufficient detection precision caused by large number of small targets, large scale change and easy complicated background interference in the plant disease and insect pest image. Disclosure of Invention This section is intended to outline some aspects of embodiments of the application and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section as well as in the description of the application and in the title of the application, which may not be used to limit the scope of the application. The present invention has been made in view of the above and/or problems occurring in the prior art for identifying pepper insect pests. Therefore, the invention aims to solve the problem that the precision is not high when detecting small target insect pests in the prior art. In order to solve the technical problems, the invention provides the following technical scheme that the pepper insect pest detection method based on the improvement YOLOv8 comprises the following steps, S1, collecting a pepper insect pest image; S2, processing the collected pepper insect pest images, and dividing the processed image data into a training set, a verification set and a test set; S3, inputting the pepper insect pest detection data set into an improved YOLOv detection network, and outputting a pepper insect pest detection result after processing by the improved YOLOv n detection network, wherein the detection result comprises insect pest type information, corresponding confidence and target boundary box positions. As a preferable scheme of the pepper insect pest detection method based on the improvement YOLOv in the invention, wherein in S2, the pretreatment step is as follows, Performing optical transformation and geometric transformation on the collected pepper insect pest image to obtain a preprocessed image; Marking the preprocessed image by using a marking tool to generate an image containing insect pest type information and marking data corresponding to the target position; And dividing the image data subjected to pretreatment and labeling into a training set, a verification set and a test set according to a preset proportion. As a preferable scheme of the pepper insect pest detection method based o