Convolutional Neural Network (CNN)of Deep Learning
- 1. Convolutional Neural Network (CNN) KIRKUK UNIVERSITY COLLEGE OF COMPUTER SCIENCE & INFORMATION TECHNOLOGY PERPETRATE BY: ALI H. AHMED SUPERVISOR DR. IDRESS HUSIEN 16 APRIL 2024
- 2. Outline What is CNN. Why do we use CNNs instead of traditional neural networks. Basic structure of CNN Input Layer Convolution Layer Pooling (Subsampling) Layers Fully Connected Layers (Dense Layers) Output layer Applications of CNN Conclusion References 2
- 3. What is CNN • 1980: Fukushima Kunihiko publishes a paper describing a prototype CNN. • Is supervised learning and type of Deep Learning a family of milt-layer neural network architecture commonly used in Computer Vision. • use three-dimensional data to for image classification and object recognition tasks. • leveraging principles from linear algebra, specifically matrix multiplication, to identify patterns within an image • The architecture of a (CNN) is analogous to that of the connectivity pattern of Neurons in the Human Brain and was inspired by the organization of the Visual Cortex 3
- 4. Why do we use CNNs instead of traditional neural networks Properties CNNs Traditional neural networks Feature extraction Automatically extracts features from data during training Requires entry of predefined manual features Working with 3D data Designed to process images and videos effectively May be inefficient in handling 3D data Specific navigation It uses pooling layers to reduce dimensionality of distinct data You may not use pooling layers, which makes them more susceptible to small changes in the location of objects Ability to deal with big data Able to effectively improve its performance with big data May be inefficient with relatively large data Hierarchical representation of features It can extract features hierarchically from data You may have difficulty representing features hierarchically 4
- 5. Basic structure of CNN 5 Input Layer Convolutional Layers Pooling (Subsampling) Layers Fully Connected Layers (Dense Layers) Output Layer
- 6. Input Layer When we enter an image, what the computer sees is a matrix of pixel values. Depending on the resolution and size of the image, the computer will see different matrices, such as a 32 x 32 x 3 matrix (3 refers to RGB values). Each digit in the matrix has a value from 0 to 255, which describes the pixel's gray level at that point. 6
- 7. Convolution Layer 7 is the core of convolutional neural networks (CNNs) and is the basic element for building them process: • Filters (Kernels): The convolution process begins by applying filters to the image. • Image scanning: The filter slides across the image, performing a raster multiplication operation with a selected area of the image. • Feature Maps: Convolution of the entire input data with a single filter produces a feature map. Each feature map highlights the presence of a specific property that the filter is designed to detect. • Multiple filters: A convolution layer can contain multiple filters, each of which produces its own feature map.
- 8. Filters (Kernels) 8 They are essential components of convolutional neural networks (CNNs) and play a pivotal role in extracting features from data Example of convolution filters: • Edge detection filters: Used to detect edges in images. • Sobel filters: A special type of edge detection filter used to determine the direction of edges. • Prewitt filters: Similar to Sobel filters, but sometimes used to reduce noise in images. • Laplacian filters: Used to detect sudden changes in pixel values. • Gaussian filters: used to reduce noise in images. • Gabor filters: Used to detect specific patterns in images, such as texture and texture.
- 9. Feature Maps 9 It is the process of passing the filter over the input image and rasterizing it to extract the features 𝑖𝑛𝑝𝑢𝑡3∗3 1 2 3 4 5 6 7 8 9 * filter2∗2 −1 1 −1 1 =Feature Maps 2 −2 2 −2 Dimensions Input (x,y) * filter (u,v)=Feature Maps (x-u+1,y-v+1) (3,3) *(2,2) = (3-2+1,3-2+1) =(2,2) (1×−1)+(2×1) (4×−1)+(5×1) =(−1)+(2)+(−4)+(5) =2
- 10. Pooling (Subsampling) Layers 10 These layers are used to reduce the spatial dimensions of incoming data while retaining the most important information. It helps in reducing the complexity of the neural network and controlling redundant separations.and characterized by parameters such as the size of the pooling window (usually 2x2 or 3x3) There are two main types of pooling layers 1. Max Pooling 2. Average Pooling
- 11. Fully Connected Layers (Dense Layers) In these layers, each node (neuron) from the previous layer is connected to all nodes in the current layer. A linear transformation is applied to the input vector using the weights matrix. A nonlinear transform (activation function) is applied to the product using a nonlinear function such as ReLU or Sigmoid. These layers are used to learn nonlinear relationships between features. They are usually used at the end of a neural network to generate the final results. 11
- 12. Output layer It is the last layer in the network that produces final predictions or classifications based on information extracted from previous layers. The function of this layer depends on the type of task being performed by the network Output layer function: Classification Object Detection Activation Functions: softmax Training and Optimization 12
- 13. Image classification applications of CNN 13 1) Object recognition: Identify objects in images (example: animals, people, cars, brands) 2) Semantic classification of images: identifying the type of image (e.g. landscapes, portraits, sports images)) 3) Verifying people's identity: Identify people in photos
- 14. Image Processing applications of CNN 14 1) Noise removal: Improve image quality by removing unwanted noise 2) Enhance images: Change image brightness, contrast, saturation, and sharpness 3) Resize Images: Resize the image without losing details
- 15. Video processing applications of CNN 15 1) Video classification: Select the video type (example: news, movies, sports) 2)Video Summarization: Create a short summary of a long video 3) Activity recognition: Identify activities occurring in the video (eg: walking, running, jumping)
- 16. Natural language processing applications of CNN 16 1) Text recognition: extracting information from texts (e.g. names of people, places, institutions) 2) Machine translation: Translating texts from one language to another 3) Text Summarization: Create a short summary of a long text
- 17. Other applications of CNN 17 1) medical diagnosis: Helping doctors diagnose diseases by analyzing medical images (such as X-rays, CT scans) 2) Stock Price Forecasting: Analyzing financial data to predict stock price movement 3) Controlling Robots: Helping robots see and understand the world around them
- 18. Conclusion CNN is a powerful tool in the field of machine learning and visual analysis. The benefits of CNN include automatically extracting features and improving classification and analysis performance. Various applications of CNN include medical imaging, image classification, face recognition, automated video analysis, and others. Future research and work in the field of CNN is essential to improve performance and discover new applications. 18
- 19. References https://gemini.google.com https://www.researchgate.net/figure/A-typical-CNN-framework-for-robotics- Raw-sensor-information-depth-image-from-an-RGB-D_fig2_311805526 https://www.analytixlabs.co.in/blog/convolutional-neural-network/ https://chat.openai.com https://www.ibm.com/topics/convolutional-neural-networks 19