Automatic Detection And Recognition Of Field Lamp Lure Insects Based On Deep Learning

Field insect monitoring is one of the most important links in the agricultural production system.The traditional field insect monitoring work relies on professional plant protection personnel to investigate,which is hard to guarantee the timeliness and accuracy,and it is also difficult to meet the current demand of plant protection forecast.The rapid technological development of computers in recent years has provided the technical basis for automated field insect monitoring.Insect recognition and classification based on images is one of the current research hotspots in the field of insect automatic monitoring.In recent years,computer vision technology based on deep learning has developed rapidly,which provide new method for the realization of automated field insect monitoring.In this paper,we studied the detection and classification of agricultural lamp-induced insect pictures by using deep learning-based object detection algorithm.Firstly,a field insects images dataset were established.Then,the field insect automatic detection and classification model is trained on the basis of images datasets.Finally,a lightweight field insect detection and identification algorithm was proposed.The main work and research results of this article are as follows:(1)A total of 3300 pictures of insects were taken with the help of agricultural lamps to attract insects by insect convergence.These images are first pre-processed,and then the insects are annotated and classified using VOTT annotation software.3000 pictures were selected as the training set,and these pictures were enhanced using affine,noise,blur,and color adjustment methods.The expanded data set contains a total of 12,000 pictures.A dataset of field lamp-induced insect images was established based on the images and labeled files.This dataset has a certain generality and can be used for various model training,providing data support for further research on the detection and classification techniques of field lampinduced insects.(2)Through theoretical overview and analysis,this paper studied the automatic recognition of agricultural lamp-induced insect pictures based on the high-performance target detection algorithm YOLOv3.Based on the Docker containerized service,the Tensorflow deep learning framework is built,and based on the field lamp lure insect image data set,a field lamp lure insect automatic detection and recognition model is proposed.The results of the test set evaluation showed that the model mean accuracy mean m AP reached 69.27%,which provides that the model is effective in detecting and identifying insects in the field.(3)According to the shortcomings of the original YOLOv3 target detection algorithm in the field light trap insect image detection and recognition tasks,and the need for low-cost detection models in agricultural production environments,a lightweight classification model for automatic insect detection in the field is proposed based on the YOLOv3 algorithm.The improved algorithm introduces deep separable convolution and generalized cross-sum ratios,using a redesigned feature extraction network.The results from the validation set show a 1.71%improvement in m AP using the improved algorithm with a significant reduction in model volume.The insect features extracted by the model were demonstrated through visual analysis,which proves that the model can effectively extract insect features from complex pictures.Finally,the improved method was compared with the classical target detection algorithms including Faster R-CNN(VGG16),YOLOv3,and YOLOv3-tiny,which proves that the improved method has advantages over the classical algorithms in terms of resource consumption,speed,and accuracy.This studies establishes a field insect image dataset and trains a lightweight insect detection model that keeps resource consumption low while maintaining low resource occupation,it also has good recognition effect,which laid the foundation for future research on field automated insect monitoring equipment.