Identification And Research Of The Ambrostoma Quadriimpressum Motschlsky Based On The Convolutional Neural Network

The ambrostoma quadriimpressum motschlsky(aqm)is a common tree species in the Northeast,and is also an important economic tree species and ornamental tree species.However,it is often infested by pests during its growth.In recent years,the use of deep learning models to identify and locate pests in images has become a research hotspot.How to accurately identify and locate the aqm image with complex background is of great significance for providing pest control.In this paper,the image of the aqm photographed in the natural environment was studied.The main contents are as follows:Comparing the effect of typical traditional target detection algorithm on the recognition and localization of the target in the aqm image,it is proposed to use the recognition method based on convolutional neural network to extract the featureless image of the image and identify the aqm.The process is divided into two steps.Firstly,the Faster R-CNN network is used to identify and locate the aqm in the image,and frame it with a rectangular frame to eliminate the influence of complex background.Secondly,the convolutional neural network was used to finely classify the framed results of Faster R-CNN,and to eliminate the interference of other kinds of pests similar to the characteristics of the aqm.Accurately identify and locate the target aqm in the image.The rough identification and positioning stage:a standard initial candidate frame length-to-width ratio that appears in the standard Faster R-CNN network recognition image is inconsistent with the morphological characteristics of the aqm itself.The adaptive clustering of the algorithm is based on the number of’ Faster R-CNN networks.Firstly,the k-means clustering algorithm combined with the BWP index is used to cluster the length and width values of the rectangular frame in the training data label of the aqm,and the cluster center point is obtained.Then,the cluster center point is used to replace the aspect ratio generated by the standard initial candidate frame,so that the initial candidate frame generated by the network is more suitable for the aqm,and the displacement of the translation during the later finishing of the rectangular frame is reduced;The size of the aqm is too large,and the size of the initial candidate frame is adjusted to achieve a more accurate frame.Fine classification stage:For the error of the improved network,the other species with similar characteristics of the aqm are identified as the target aqm and framed.It is proposed to use the convolutional neural network for Faster R-CNN’s framed results are carefully classified.Firstly,design and build a convolutional neural network for meticulous classification.For the problem that the size of the output rectangle of the Faster R-CNN network is different,the final output feature vector length can not be classified,and the last layer of the network can be replaced.For the network that can normalize the features;secondly,multi-scale and sharpen the data of the training meticulous convolutional neural network,and train the convolutional neural network;again,extract the Faster R-CNN results The area in the middle rectangle is preprocessed to increase the difference between the target and the interferer and the background.Finally,the network recognition result is re-marked back to the original inage to obtain the final recognition and positioning result.The image of the aqm adjacent to the leaf gap,the two aqms,and the similar characteristics of the aqm are identified,and the algorithm is compared with the other two current mainstream deep learning.The effect of the target detection algorithm to identify the image of the aqm is compared and analyzed.which proves the effectiveness of the algorithm.