Design And Application Of Multi Convolution Neural Network Collaboration Model

Computer vision is an important branch in the field of artificial intelligence,and image recognition is an important research direction in the field of computer vision.In recent years,with the rapid development of deep learning technology,image recognition technology has also made significant progress.From the early Alex Net to VGG-16 to Res Net101,the recognition accuracy of image recognition models based on deep neural networks is getting higher and higher,but the stability of the model still needs to be improved,especially in the face of complex lighting and multiple shooting angles.The recognition accuracy of an object is also very different.Therefore,for the problem of image recognition under complex lighting and multiple shooting angles,this paper combines a multi-intelligent system and a deep neural network to build a multi-convolutional neural network collaboration model based on deep neural networks and applies it to images under complex lighting Recognition,pig face recognition and tire X-ray defect detection.Compared with traditional deep learning models,the model proposed in this paper has better stability for image recognition under complex lighting and multiple shooting angles.The main content of this article includes the following three parts:(1)Aiming at the problem of image recognition under complex lighting,a method of image classification based on a multi-convolutional neural network collaborative model is proposed based on the idea of multi-intelligent systems and a variety of traditional digital image processing feature extraction methods.This method first preprocesses the data set through feature extraction and clustering,which reduces the impact of complex lighting on the image.The multi-convolutional neural network collaboration model enables the corresponding convolutional neural network model to recognize the image according to the PCA feature vector,LBP feature vector,and HOG feature vector of the input image.The multi-convolutional neural network collaboration model integrates each convolutional neural network.The output results are fused to obtain the final image detection results.Experiments show that the multi-convolution neural network collaborative model has better classification accuracy in image data sets of complex lighting environments.(2)Aiming at the problem of pig face recognition under complex lighting and multi-angle shooting,combined with semantic segmentation,a multi-convolution neural network collaborative pig face recognition algorithm based on semantic segmentation was proposed.Pig face recognition is to identify the pig's identity through the image of the pig.Because the environment of the pig house is relatively dark,the images are often taken with uneven lighting conditions,and the pig is not always facing the camera,and the pig face recognition has always been difficult.Due to the large differences in the features of various parts of the pig,the recognition of whole pigs cannot obtain satisfactory results.This paper builds a multi-convolution neural network collaborative model based on semantic segmentation.The model can recognize whole pig images and can also be semantically segmented.Find and recognize the pig's head and tail in the input image,and the model fuses the recognition results of the head,tail and whole pig images to get the final detection result.Through experiments,it can be proved that compared with traditional deep learning models,the model proposed in this paper has better detection effect on pig face recognition under complex lighting and multiple angles.(3)The X-ray image of the tire is taken by an X-ray machine to obtain the internal structure of the tire.When the tire contains different defects,the internal structure of the tire will also change accordingly.At present,the main detection method is to manually identify the defect,which cannot always guarantee the accuracy.Because different types of tires have different internal structures,the traditional target detection model will recognize the X-ray image of a certain type of tire as an image containing a defect during detection.In view of the above problems,this paper applies the idea of multi-convolution neural network cooperation model to Faster R-CNN,and proposes a cooperative X-ray defect detection algorithm based on the target detection model.Each target detection model corresponds to a type of tire.When a specific type of tire detects a specific type of defect,the image is input into a discrimination module to obtain the final detection result.In the tire X-ray image defect data set,the proposed method has better accuracy and higher recall rate.