Research On Detection Of Larch Caterpillar Pests Based On Deep Learning

The larch caterpillar(Dendrolimus superans Butler)belongs to the order Lepidoptera,and it is widely distributed in Inner Mongolia,the three northeastern provinces and the northern region of Xinjiang.The traditional detection methods are mainly divided into manual detection,remote sensing satellite monitoring and sensor technology monitoring,etc.The disadvantages are that the detection cycle is long,the detection accuracy is low,and it requires a lot of manpower and material resources.It cannot respond to forest disturbance in time and prevent it in time.In response to these problems,this paper takes larch caterpillar pest trees in forest farms in Daxing’anling area as the research object,and proposes a larch caterpillar pest detection model based on deep convolutional network.The experiment proves that the accuracy rate of this detection method can reach 97.50%,and a real-time monitoring system for small equipment is built based on this detection model.On this basis,this paper simplifies the model framework to finally realize the real-time detection of larch caterpillar pest trees.The main research content of this paper is as follows:(1)The characteristic information of trees damaged by larch caterpillars was clarified,and a data set of larch caterpillar pests was constructed.First,after the research on the sample plots,the drones were used to collect the aerial image data of two experimental forest farms in the Greater Khingan Mountains,and preprocessing methods such as cropping and enhancement were performed on the images.Secondly,the LabelImg open source project was used to calibrate the pest targets and design The dead tree target in the forest area is calibrated as the interference feature,and the dead tree and pest characteristics are stitched together according to the Mosaic data enhancement method.Finally,after data amplification,2900 training images,360 test images and verification images of the larch caterpillar pest data set are obtained.360 pictures.(2)In view of the low efficiency of traditional larch caterpillar pest tree detection methods,this paper proposes a larch caterpillar pest detection model based on the one-stage detection method YOLO v4 and the two-stage detection method Faster R-CNN.The two detection methods are trained respectively,and the generated models are deployed to the computer for testing and the results are compared and analyzed.The test results show that the Faster R-CNN model has a pest detection accuracy of 96.53%,higher than 94.10% of the YOLO v4 model,and the detection speed of the Faster R-CNN model in the detection of a single aerial photo is 0.175 seconds per frame,which is much lower than that of the YOLO v4 model.The YOLO v4 model is 0.065 seconds per frame,but the real-time detection speed of the YOLO v4 model is only 1.72f/s,which does not meet the requirements for being mounted on the edge computing platform.(3)The detection accuracy of the Faster R-CNN model is slightly higher than that of the YOLO v4 model,but real-time detection cannot be achieved under existing equipment.Therefore,the training cost for the two-stage network model is high,and the detection speed of the edge computing equipment carried by the UAV is high.For low-level problems,a real-time detection method for affected trees based on the improved YOLO v4 model is proposed.By optimizing the model structure,the recognition accuracy and detection speed of trees damaged by larch caterpillars are improved,and the improved network model is lightweight.The results show that the accuracy of pest detection is reduced by 0.57% compared with the original network,the overall average detection accuracy is increased by 0.28%,and the real-time detection speed is increased by 30.28 f/s.While ensuring the accuracy,the detection speed is greatly improved,and the cost of model application is reduced.(4)Developed a recognition system for larch caterpillar pest detection,and developed a larch caterpillar pest recognition model based on airborne edge computing equipment and verified it.Based on the improved YOLO v4 model larch caterpillar pest detection method,develop a larch caterpillar pest recognition system,call Python to call the PyTorch-based YOLO v4 model,and complete the target detection of larch caterpillar pest trees in UAV aerial photos And counting,calling the device camera to detect and count the pest targets in the surrounding environment.To sum up,through the use of UAV aerial photography technology and the development of edge computing equipment,this research has completed the above research work,and realized the target detection function through ground testing.This study provides forest managers with quantitative information on the location of pests,provides an objective and accurate basis for the assessment of damage to trees affected by larch caterpillars,and realizes pest monitoring based on modern remote sensing technology.In addition,the process is not only based on the larch caterpillar pest system It can also be widely used in other fields of UAV by following the steps of this research process,which has certain reference value.