Research On Multi-objective Image Segmentation And Recognition

Image segmentation and recognition is one of the important research directions in the field of machine learning.The main task is to distinguish the target from non-target in image data or video data.If there is a target,it is necessary to determine the location of the target and classify and recognize it.In this paper,the improved Faster R-CNN algorithm is used to segment and recognize VOC data sets and self-made office supplies data sets.The advantages and disadvantages of the original algorithm and the proposed algorithm are analyzed and compared.The improved algorithm is proposed by combining feature fusion with AlexNet model and VGG model framework.In this paper,three methods of multi-objective image segmentation and recognition are proposed.Aiming at the problems of small target,occluded or complex background image segmentation and recognition results and low efficiency of classical Faster R-CNN method,feature fusion is introduced,and an improved Faster R-CNN method based on feature fusion is proposed.The main idea is to fuse the features generated by mapping candidate regions to different convolution layers in convolution neural network,stack the input feature maps in the dimension,get the fixed size feature vectors,and send them to the full connection layer for classification and boundary box regression.The experimental results show that this method can improve the segmentation and recognition results of multi-target images compared with the classical Faster R-CNN method.In the network framework of Faster R-CNN method,AlexNet model of convolutional neural network can not adapt to large gradient input in training because of using ReLU activation function,which results in low segmentation and recognition accuracy.An improved Faster R-CNN method based on AlexNet is proposed.This method uses Leaky ReLU activation function to reduce the occurrence of silent neurons,thus allowing gradient learning.The experimental results show that the improved AlexNet model improves the accuracy of multi-objective image segmentation and recognition by weighting the negative data,and greatly improves the efficiency.In view of the problem of gradient disappearance and gradient explosion in VGG model of convolutional neural network,an improved Faster R-CNN method based on VGG is proposed and improved in VGG16 and VGG19 network models respectively.By changing the structure of VGG network and introducing Leaky ReLU activation function to adjust parameters,the problem of gradient explosion can be solved.The experimental results show that the improved VGG model has more activation functions and more convolution kernels,which makes the decision function more discriminative,effectively improves the classification and recognition effect,and further improves the accuracy of Multi-target Image Segmentation and recognition.