WEBVTT 00:02.120 --> 00:06.320 Hello and welcome to the Street Odille on Convolutional Neural Network. 00:06.560 --> 00:10.800 Let us begin a convolutional neural network. 00:10.940 --> 00:16.310 The type of neural network that is mostly used for the image processing problems. 00:17.690 --> 00:24.290 The advancements in computer vision and deep learning has been constructed and preffered with convolutional 00:24.290 --> 00:25.430 neural network. 00:28.450 --> 00:35.800 There are multiple applications of convolutional neural network, like a major recognition, video recognition, 00:36.040 --> 00:43.060 image analysis, image classification, Milioti creation recommendations, esteems natural language 00:43.060 --> 00:43.840 processing. 00:44.110 --> 00:46.450 And this list goes on and on. 00:49.570 --> 00:55.470 To understand the convolutional neural network, we have to understand the different layers in that. 00:56.050 --> 01:03.490 These are the five main layers in convolutional neural network input layer convolution layer at a layer, 01:03.790 --> 01:07.030 pooling layer and fully connected layer. 01:07.750 --> 01:11.010 Let us understand these five layers, one by one in detail. 01:14.240 --> 01:19.220 This is the pictorial representation of delayers in convolutional neural network. 01:20.000 --> 01:26.870 As you can see here, first VLT input layer, then convolution, plus add a little layer after that 01:26.870 --> 01:27.800 fulling layer. 01:28.220 --> 01:30.470 Then again, convolution pessary, Lou. 01:30.590 --> 01:38.060 And pulling the combination of these three layers, that is convolution plus RELU plus pulling layer 01:38.150 --> 01:39.710 is collected in layers. 01:40.340 --> 01:46.940 After that, we have to flatten the data and provide the input to fully connected layer to the output 01:46.940 --> 01:48.350 of fully connected layer. 01:48.410 --> 01:51.590 We have to apply activation function here. 01:51.600 --> 01:55.550 We have to apply softmax or sigmoid activation function. 01:56.300 --> 01:58.850 As of now, just remember the name of layers. 01:59.000 --> 02:00.050 Nothing is. 02:03.120 --> 02:09.660 So this is the first layer in convolutional neural network that is input layer, the input layer the 02:09.680 --> 02:13.720 horse, the raw input of image with very height and depth. 02:15.360 --> 02:21.730 The convolution layer computes the output volume, and to do that, it calculates the dot product between 02:21.730 --> 02:23.230 filter and image. 02:24.010 --> 02:26.890 Let us understand the convolution layer in detail. 02:30.050 --> 02:35.730 The convolution layer is the first layer to extract the features from an input image. 02:37.190 --> 02:45.470 It prevents the relationship between pig cells by learning the image features convolution is a mathematical 02:45.530 --> 02:53.490 operation that takes to input flat input is the image matrix and the second input is built. 02:54.380 --> 02:57.530 So theoretically, this is all about the convolution layer. 02:57.770 --> 03:00.400 Let us understand the mathematical operation. 03:03.470 --> 03:08.290 To simplify the calculations, we take an image of the save for bifold. 03:08.750 --> 03:11.530 And after that, we help filter to by two. 03:13.050 --> 03:19.360 After the multiplication of image and filter, we will have the output of essays three by three. 03:19.980 --> 03:21.270 Let us understand these. 03:22.800 --> 03:26.580 First, we will apply this filter at the cordon it off this image. 03:27.360 --> 03:37.220 Now multiply one into one one zero into one zero zero in two zero zero and one two one one. 03:37.590 --> 03:44.580 So one plus one two after that, move this filter towards the right. 03:46.780 --> 03:55.810 No, again, same calculations, one into one one zero in two zero zero zero and two one zero and one 03:55.810 --> 03:57.370 in two zero zero. 03:57.850 --> 04:00.790 Add all these values that these one. 04:01.840 --> 04:06.610 In the same way, we help to do all the calculations like these. 04:13.340 --> 04:17.600 And at the end, we will have the output of size three by three. 04:18.170 --> 04:20.990 So this is how the convolution layer works. 04:24.040 --> 04:30.940 If the input image is a squared sized image and input length is N garner, length is key. 04:31.180 --> 04:33.260 Then output length of that image. 04:33.280 --> 04:35.890 D and minus K plus one. 04:37.510 --> 04:45.160 And if the input image is not this side's image, then output, height of that image is equal to input 04:45.160 --> 04:53.650 height minus col, height plus vun and output width is equal to input with minus Cardinal Vade plus 04:53.650 --> 04:54.250 one. 04:57.230 --> 05:04.270 Bill, now in this story, you'll be held under two two CNN layers that these input layer and convolution 05:04.270 --> 05:08.540 layer, we will understand the remaining three layers in next tutorial. 05:08.790 --> 05:10.600 Till then, happy learning.