| Fruit fly pests seriously affect the quality and safety of various fruit and vegetable crops.Many fruit farmers lack sufficient knowledge of the levels of pest occurrence,leading to the transitional use of pesticides,resulting in environmental pollution and changes in crop quality.At present,the trapping equipment commonly used by fruit farmers can only achieve the role of pest extermination,which requires frequent manual replacement and cannot completely solve the problem of pest infestation;the equipment used by large farms is bulky,expensive and high maintenance costs,and cannot be truly applied to the grassroots.Traditional pest monitoring and early warning in agriculture and forestry use four methods: manual counting,statistical counting of sound signals,infrared sensor counting,and image processing,but all of these methods have problems such as low recognition effect and counting accuracy,and are easily affected by the external environment.Therefore,this paper takes the Bactrocera cucurbitae as the research object and uses artificial intelligence and agricultural technology together to design a live pest detection and counting system for orchards containing trapping devices to help researchers and fruit farmers grasp pest data in a timely manner,thus reducing the use of artificial and pesticides and thus achieving scientific warning and control of pests.The main research components are as follows:(1)For the common market trapping bottle,trapping ball,sticky cardboard and other large three-dimensional trapping device shortcomings,this paper designed a set of trapping function and access to live pest image function as one trapping device,the device is not only convenient and portable,affordable,easy to install,small footprint,and has the Internet communication function,the image data obtained can be uploaded in real time for subsequent pest detection and quantity statistics.(2)By adjusting the relevant parameters and designing the program,some improvements were made to the Hough circle detection algorithm of OpenCV,so that it can stably and continuously detect the entry and exit holes of the trapping bottle device and get the position information for subsequent counting of the number of pests entering the holes.(3)The dense optical flow algorithm Optical flow,the semantic segmentation algorithm U-Net,and the target detection algorithm YOLOv5 are proposed and designed for the detection and counting of live pest planes,and the accuracy of these three detection and counting methods are compared and analyzed,and finally the YOLOv5 algorithm is selected as the pest detection and counting method.To further improve the accuracy of YOLOv5 network model for pest detection,this paper used the Mosaic data enhancement technique to preprocess the data,conducted a cluster analysis of the Bactrocera cucurbitae image dataset to obtain more suitable anchor boxes for the Bactrocera cucurbitae detection model,and did a comparison experiment of adding four different types of attention mechanisms,SE,CBAM,CA,and ECA,to the YOLOv5 network model architecture respectively.(4)The algorithm design related to the pest location information and the location information of the entrance and exit of the trapping bottle was carried out to achieve the effect of plane detection counting of live pests and counting the number of pests entering the trapping bottle.Applying the method proposed in this paper,three sample videos were tested and analyzed,and the results showed that the average accuracy of detecting and counting Bactrocera cucurbitae on the surface of the trap bottle reached 93.5%,and the accuracy of counting Bactrocera cucurbitae entering the trap bottle reached 94.3%.The experimental results show that the method designed in this paper can accurately detect and count the living Bactrocera cucurbitae,which provides a new information analysis method for pest control.Its application can improve the efficiency of scientific research on pest control,and has good prospects for application in plant protection,forest pests and diseases,and pest control. |
PDF Full Text Request