Automated Flower Prediction Using convolutional Neural Network
The flower prediction is very important in the agriculture and it is also mainly useful in botany department where the flower classification is very important in finding the plant. The flower prediction can be done with the help of deep learning. The deep learning is the type of machine learning which is used to classify the image. The deep learning helps to many researchers who are going to perform research by collecting, organizing and classifying the information. In deep learning we are going to use the convolutional neural network. The convolutional neural network contains different types of architecture and different types of layers. In this paper we will discuss briefly about classification of flowers using the convolutional neural network.
Keywords: Deep learning, convolutional neural network,
The artificial network is the very powerful technology in which making the machine to think. In recent years the neural networks were very popular and now a day up to the year 2010 the machine learning was very popular. The present generation was mainly focused on the deep learning network. It performs the analysis very accurately at very deep. It helps many researchers and data scientist to perform their task very easily. It is somewhat like neural networks but performs the computations very deeply. It was very advanced. It is a subset of machine learning. It also works like a human brain. It also helps to solve the complex problems very easily. It has layers exactly like exactly neural networks. It has inner layer, hidden layer and output layers.
Fig 1: AI Technics
The deep learning has more number of parameters and can solve many complex problems. There are four fundamental architectures. They are
- Unsupervised pre-trained networks
- Convolution neural networks
- Recurrent neural network
- Recursive neural network
The deep learning is very useful for data scientist who is collecting, analysing and interpreting large amounts of data. It is very useful for classifying the image. The process of deep learning is very fast and efficient.
Fig 2: Deep Learning Process
Step 1: The researcher should analyse the problem and how to find the solution and in what the problem can be solved. In all aspects the researchers need to analyse the problem and select the methodology which it best suited for execution.
Step2: The next work what the research has to do is search the relevant data which is used to analyse the problem.
Step3: select the deep learning algorithm and execute the process.
Step4: Train the algorithm with labeled data.
Step5: Test the algorithm with unlabelled data.
These are steps for performing the deep learning process. It is mainly used for image classification. The image classification is used in many areas for research purpose. The deep learning classifies images layer by layer. Each layer sends some information to the other and then integrates together. So it performs all analysis very deeply.
The deep learning can perform any image classification on supervised data or unsupervised data etc. it is very useful for image recognition. It is also used in all areas.
Fig 3: Deep Learning Steps
Objective of the Course Work
The main problem in this paper is to classify the flowers according to their plants. The classification of flowers is very important. It is used in botany department because it was very essential to classify the flowers according to their plants. It helps to identify the diseases of plants. The image classification is done with the help of convolutional neural network. The convolutional neural network is very useful for analysing the images. The image recognition can be done very easily. The initial processes are first we should perform the image pre-processing and afterwards the feature extraction is performed. After performing feature extraction there are different types of architecture are there in convolutional neural network. In that we can select any one from those architectures which is used to classify the image. Finally the flowers are classified according to their types.
The author states that (Katarina Mitrovie, 2019) the flower classification is very important in agriculture and in botany department and it is also very useful in many areas which is used in entertainment also. The image classification is performed by using the convolutional neural network. It was very easy to perform image recognition in convolutional neural network. The author selects the AlexNet and LeNet models for image classification. The author also focuses on sigmoid uniform function which is used to assign weights in the network. There are different types of models in convolutional neural network. The author finally calculates the performance metrics for all three models and confusion square matrices is also used for calculating the performances. The image is finally classified according to the labels.
The author proposes that (An Tien Vo, 2017) proposes that the image classification is the very important and complex task in computer environment. The classification of images is done with the help of convolutional neural network. The convolutional neural network works well with for the classification of images. The image classification is very important in many fields. The author proposed the new algorithm for the image classification.
The author states that (Baizel Kurian Varghese, 2020) is stated that the trees are grown in our country differently. To classify the trees by using the leaves are the main objective of this paper. This paper focuses on convolutional neural network. It is one type of neural network. It is mainly based on classification of leaves. The leaves are grouped according to their types. Most of the leaves look same. The convolutional neural network has different types of architecture which helps to classify the nodes.
The author states that (Philipe A. Dias, 2018) the fruit production is the very important measure in certain department. To perform that process we need the count of how flowers that present in the plant. Based on that, the neural networks are used. The convolutional neural network is used to classify the images.
The author focuses on the automatic classification of plants. In this paper the author tells that (Hulya Yalcin, 2015) plant classification is very important in developing the agriculture. Nowadays the agriculture is developed to meet the modern way of producing crops. The agriculture is now moving towards IOT that is smart way of agriculture. The plant classification is very important in identifying species. It can be done with the help of convolutional neural network.
The author states that (Jimmy Lei Ba, 2015) the text is generated automatically to the visual images. The semantic models are used to classify the models. The author used the convolutional neural network to recognize the image objects. Initially the weights are assigned to the node in the convolutional neural network. By using this convolutional neural network the image is recognized easily. The layers contain different types of architecture which helps to classify the objects.
The author states that (Francesco Visin, 2016) the recurrent neural networks are used more. The author recognizes pictures with the help of recurrent neural network with advanced features. It has many different numbers of layers which helps to classify the image. The image classification is very important for business people also. It works and integrated with many layers. Initially the layers are trained with the datasets. After training the layer the images are classified according to the class labels.
The author states (Thanh-Binh Do, 2017) that the classifying of organs of plants is very important in botany. In this paper they have used fusion technique which helps to classify the organs. They have used the convolutional neural networks and they discussed about deep learning technique. The deep learning is used to give very accurate results. The plant identification is also very important in all cases. The plant organs are classified according to the species of plants. The author also used the various fusion techniques which help to increase the accuracy of identifying the plant.
The author states that (Vidit Kumar, 2020) content based image retrieval is the most important technique which is the given query. The main work of database is to display the images related to the question. The major task of retrieving image is displaying the similar images based on the question. The author proposed that the use of convolutional neural network is very useful for image prediction, classification and segmentation. The convolutional networks are mostly used for this purpose. The author develops the proposed model to reach the superior output. There are number of layers in the convolutional neural network. They are input layer, hidden layer and output layer. The deep learning is very much useful in classifying the image. It is very powerful for image classification.
An overview of coursework
The proposed model focuses on classifying the images based on the species of the flower. The flowers are classified due to many reasons. The flowers classification is very important in botany department. The data set is taken from the Kaggle website. The species of flowers are classified even to identify the disease affected plants. The usage is very high. The deep learning is very important in the artificial intelligence. The artificial intelligence is trying to make the machine visible like the humans to the real world. The deep learning has different types of layers.
Among them we are going to use convolutional neural network. The neural networks are exactly it looks like brain in the human being which process each results to another node with weights. The neural networks are very useful. It has numerous numbers of nodes and processing elements which helps to process the information. The Kaggle website contains nearly 5000 species which help to classify the flowers. This classification will be used by many researchers and many biological departments. The image pre-processing is the important step in classifying the image. The image may be sometimes blurred, no sharp edges and looks so dull etc. These can be rectified by using the image pre-processing process.
The image pre-processing is done to enhance the image quality by using some algorithms. It also helps to get some useful information. The image is composed of finite number of elements which has some value at the particular location. The image consists of many pixels which decide the resolution of the image. The resolution of the image may be dull of bright according to the pixels positions. There are different types of images. They are binary image, colour image, black and white image are the different types of images.
The image can also be represented in two dimensional matrixes. There are different steps in image pre-processing. They are image acquisition, image enhancement, image compression and image restoration. These are the different types of process in image pre-processing. After performing the image pre-processing process the image is sent to the neural networks which further it is used to classify the image.
Fig 4: Steps in Image pre-processing
The above diagram shows the image pre-processing process. They are the first input image is downloaded and the pre-processing work is done and feature extraction is performed. The feature extraction maps the values with the original image. It will keep only the needed values. Finally the classification is done with the help of classifier.
It is used to reduce the dimensionality of the image. It is also used to select number of variables in the features which helps to reduce the information about the images. It helps to describe the data completely about the original data. It also helps to perform the process very quickly.
Convolutional Neural Network
The convolutional neural network is the one type of deep learning. The deep learning has many different methods. The convolutional neural network helps to classify the image very easily. There are different types of architecture and different types of layers in convolutional neural network. The convolutional neural network will share their parameters. The convolutional neural network also contains so many filters. The convolutional neural network is the very powerful tool in classifying the image pre-processing process. It also helps to solve all the problems very easily. It also gives maximum accuracy. This neural network is mainly used for image classification. The network has series of layers to perform maximum level of accuracy
It has so many filters which help to filter the image and it has simple receptive field and it also performs full depth of the input.
It reduces the dimensions of the image. It also reduces the number of parameters and number of computation performed in the network. The pooling layer finally summaries the feature map which was produced by the convolutional layer. This is another important layer in convolutional neural network. It is the layer to select the needed sources. It also reduces number of computations.
Activation function Layer
There are different types of activation function are there. The activation function helps to perform some operations on the neural network. There are different types of activation function are there. The different types of activation function are sigmoid, linear function and Relu function etc. These are the different types of activation function in convolutional layer.
Fully Connected Layer
It is an important component in convolutional neural network which helps to recognize and classify the images. The output generated from this layer is fragmented into small and it is given as input to the fully connected layer which makes an important decisions.
There are different types of architecture there in convolutional neural network. They are as follows.
The classic network architecture are
The modern architectures are as follows
These are the different types of architectures which help to execute the simulation.
It was developed in 1998 to identity the handwritten digits for zip code recognition.
Each layer has filters and these filters will reduce the computation.
It was developed in 2012 to compete in the ImageNet competition. This model is larger and main success of this model is it mainly focuses on deep learning.
It was introduced in 2014 and at the introduction of this model it mainly focuses on deep learning.
The researchers at 2014 have introduced the Inception which was the leading top model during this time. There are different types of filters at each step. They are 1*1, 2*2, 5*5 are the different types of filters which helps to extract the features. It also helps to extract the features at large scale. The overall network throughput was improved by introducing two output layers. The result was very much good. The deeper the network will generate more cells in this model.
The ResNet is very modern approach which develops the network at very deeper. It gives the very good performance. It also helps to solve many complex tasks. The addition of more layers in the network will increase the accuracy. For any model the accuracy and the speed of computation is very important. There are many new models which help to increase the accuracy of the model. The models are very useful and helpful to generate the model to classify the image. These are the some of the architecture which helps to support the network to classify the image. The image classification is the very important and necessary process which is very helpful in many ways. This is the architecture in which h we are going to implement.
The data set are collected from Kaggle. The flower classification is very important in many areas. Based on the flower classification it also helps to identity the plants diseases. The flower prediction is very useful in business and also in botany. The flowers are classified according to their features. The simulation contains three phases. They are training phase, testing phase and validating phase. These phases are used to train the model with example. After training only the image can be tested and validated. There are nearly 5000 sample species for the flowers. These flowers should be classified according to their species. In this five species of flowers are classified.
The model is first trained with the network. The network then undergoes further process. The author uses the confusion matrix to analyse the result of network.
The developed algorithm should be analysed to know the accuracy. The algorithm should be developed to achieve good accuracy and maximum throughput. The theoretical model is developed for the five species of flowers and testing, training and validation process are performed. The correctly classified and incorrectly classified calculations are performed. The accuracy of the algorithm in training phase is 96.6 and in validation phase are 96.54 and in testing phase are 95.82. The maximum level of accuracy is achieved.
The following screen shots will show the selection of species of flowers for training, testing and validation.
Fig 5: Sample Daisy Flower
Fig 6: Sample Dandelion Flower
Fig: 7 Sample Rose Flower
Fig 8: Sample Sunflower Image
The above screen shot shows the selection of several flowers and the accuracy is calculated for each training phase, testing phase and validation phase. Hence the maximum level of accuracy is maintained. The developed model should have good accuracy. The developed algorithm should undergo three steps. They are training, validation and testing. The accuracy is calculated at all levels. The accuracy in confusion matrix for each flower is calculated. The accuracy for daisy flower is 97.0%, for dandelion is 96.5%, for rose is 96.5%, for sunflower is 95.2, for tulips is 96.1.
The confusion matrix is used to test the performance of the classification. It is also called as error matrix. The matrix is formed by calculating by the number of correct predictions and incorrect predictions. It also shows that the way in which classification model is confused. It also helps to find the errors in the generated model. The algorithm will capture the true positive, true negative, false positive and false negative. In this way the performance of the algorithm is calculated.
The flower prediction is done in this paper to help many business and the needed departments. The flower prediction is very important in science. The flower prediction also helps to find the plant diseases. The deep learning model is selected to perform this process. The deep learning analysis the problem very deeply and also helps to find solution for the complex problems. The deep is very big area where many problems can be solved by using this algorithm. It is one of the famous technologies today. In deep learning the convolutional neural network is used. It is used to classify the image. It also helps to find the correct results with high accuracy.
There are different types of architectures in convolutional network. There are also different types of layers in convolutional neural network. These layers will help to find the result very accurately. The neural network is first trained and validated and finally it is tested. The accuracy of these three levels is stated above. Hence maximum level of accuracy is achieved. The simulation is performed for five sample flowers. After that the training, testing and validating process are performed. Finally the confusion matrix is calculated and the performance of calculation is analysed. These are the ways in which neural network model is developed. Like convolutional neural network the RNN is another neural network in deep learning which is used in many areas.
The image classification is done successfully and the confusion matrix calculation is also done. The confusion matrix calculates the true positive, true negative, false positive and false negative. Based on this four captured value the image classification is performed.
An Tien Vo, H. S. (2017). Advertisement Image Classification Using Convolutonal Neural Network. 9th International Conference on Knowledge and Systems Engineering , 6.
Baizel Kurian Varghese, A. A. (2020). Infoplant: Plant Recognition using Convolutional Neural Networks. Fourth International Conference on Computing Methodologies and Communication, 8.
Francesco Visin, M. C. (2016). ReSeg: A Recurrent Neural Networks Based Model for Semantic Segmentation. IEEE Conference on Computer Vision and Pattern Recognition Workshops, 8.
Hulya Yalcin, S. R. (2015). Plant Classification using Convolutional Neural Networks. IEEE, 5.
Jimmy Lei Ba, K. S. (2015). predicting Deep-Zero Sot Convolutional Neural Networks using Textual Description. IEEE International Conference on Computer Vision, 6.
Katarina Mitrovie, D. M. (2019). Flower classification with Convolutional Neural Networks. 23rd Internationl Conference on System, Theory, Control and Computing, 6.
Philipe A. Dias, A. T. (2018). Multi-species fruit flower detection using a refined semantic segmentation network. IEEE Robotics and Automation Letters, 8.
Thanh-Binh Do, H.-H. N.-T.-N.-T.-H.-L. (2017). Plant Identification using score based fusion of multi organ images. 9th International Conference on Knowledge and Systems Engineering, 6.
Vidit Kumar, V. T. (2020). Content based Fine Grained Image Retrieval using Convolutional Neural Network. 7th International Conference on Signal Processing and Integrated Networks, 6.
Know more about UniqueSubmission’s other writing services: