Convolutional Neural Networks (CNN)
Unlike the standard two dimensional array, these neural networks are a three dimensional ordering of the neurons. The very first layer is termed as the convolutional layer. Every neuron that is present in this layer operates on the data that comes from the small part of the visual field.
All the input attributes are taken in a group wise manner similar to a filter. The model interprets the images in segments and is able to process these operations a number of times so as to complete the entire image processing. The main function of processing involves the transformation of the image from the RGB or HSI scale to the greyscale. Furthermore, any modification in the values of the pixels help in identifying the edges and then the images are grouped into distinct categories.
The propagation followed is in one direction in which the CNN model may comprise one or more convolutional layers. This is followed by pooling whose operation is to lessen the spatial size of the information representation to decrease the number of the measurable factors and computations in our model. Then the output of the convolutional layers is given to a fully connected neural network that categorizes the images as shown in the below figure. We use the filters to draw out the specific parts of the image. In the multi-layer perceptron, these inputs are multiplied with their respective weights and fed to the activation function. The convolution uses the rectified linear unit function and the multi-layer perceptron makes the use of non linear activation functions followed by the softmax function. Even though the design of the convolution network might be a bit complicated, it uses less parameters to train itself compared to a fully connected layer.
Applications of Convolutional Neural Network
- Deciphering the Facial Recognition: We follow certain steps to decode facial recognition along with some major components. The first step involves the identification of the face present in the image. The second step focuses on the features of the face excluding the external factors like pose, light, disturbance, angle, etc. After the identification of the unique features, we compare the collected information with the existing training data in the database to complement the face with its name.
- Comprehension of the Climate: They help in playing a very important role in order to fight against the climatic changes. It is able to analyse the reasons on why we are able to see such drastic changes in the climate and what can be done so as to damp this effect. This helps in providing the social as well as scientific insights.
- Medical Image Computation: Medical images comprise a lot of data evaluation that stimulates from the original image recognition. The classification of medical images that are done by the CNN model helps in pointing out the inconsistencies found on the X-rays or the MRI images through a lot more precision as compared with the human eye. The model is able to find the differences between the given sequence of images, thus providing the basis for further predictive analysis.
