International Research Journal of Engineering and Technology (IRJET) Volume: 09 Issue: 08 | Aug 2022
www.irjet.net
e-ISSN: 2395-0056 p-ISSN: 2395-0072
RICE INSECTS CLASSIFICATION USIING TRANSFER LEARNING AND CNN Anindita Rath, Rasmita Routray, and Dr.Prof Sribasta Behera
Odisha University of Technology and Research, Bhubaneswar ----------------------------------------------------------------------***-----------------------------------------------------------------------A. CNN Abstract-- One of the biggest challenges to food security worldwide is insect pest attacks. Entomology has had many applications in many biological domains (i.e. insect counting as a biodiversity index). To meet a growing biological demand and to compensate a decreasing workforce amount, automated entomology has been around for decades. This challenge has been tackled by computer scientists as well as by biologists themselves. This thesis investigates the ways to classify different insect pests using various techniques. Generally these approaches undergo feature extraction, classification methods on the tested datasets. Although various techniques were proposed, transfer learning based methods are limited in literature which addresses the aforesaid problem. Presently two transfers learning based on CNN architectures were performed. The pre trained CNN models such as Alexnet and VGG16 were selected for our experiments. From the experimental results, it is observed that transfer learning can address this classification with minimal training requirements and the Alexnet is more effective in comparison to the VGG16 CNN model in terms of accuracy.
Index Terms-- CNN, Deep learning, Transfer learning, VGG16, Alexnet Classification.
Agriculture field is one of the central points that are identified with social steadiness and monetary improvement. Nonetheless, a few hundred distinct types of insects are discovered connected with put away grains and their items, and insects that assault our stores of oat sustenance constitute a standout amongst the most genuine dangers to our development. The manual grouping of such creepy crawly bothers in paddy fields can be tedious and requires generous specialized ability. The undertaking turns out to be all the more difficult when bug irritations are to be perceived from still pictures utilizing a mechanized framework. Pictures of one bug vermin might be taken from various perspectives, messed foundation, or may endure change, for example, revolution, commotion, and so forth.
|
Impact Factor value: 7.529
A CNN architecture is formed by a stack of distinct layers that transform the input volume into an output volume (e.g. holding the class scores) through a differentiable function. A few distinct types of layers are commonly used and they are convolutional layer, pooling layer, ReLU activation layer and fully connected layer [2]. Feature learning It is the first part of the architecture which receives the image input and extracts important features. These important features are extracted using convolutional layers. The pooling layers are used for reducing the spatial dimensionality of the representation, saving a lot of computational power. and also reducing the risk of over fitting. Classification
I. INTRODUCTION
© 2022, IRJET
In deep learning, a convolutional neural network (CNN or ConvNet) is a class of deep, feed-forward artificial neural networks, most commonly applied to analyzing visual imagery. CNNs have emerged from the study of brain’s visual cortex. These type of deep neural nets have been used in image recognition since 1980s [9].
|
As the name suggests, in this part the input is the extracted features from the feature learning part which are used for training the fully connected layers for classification. The final fully connected usually outputs the prediction of the image. B. Alex Deep Neural Network Alex Net’s design has 60,000 total parameters spread over eight layers. Three fully linked layers and five convolutional layers make up these eight layers. Other significant advancements include the use of multiple GPUs for training and the use of enhanced versions such as flipping, scaling, and noising of the pictures for training. Furthermore, the network employed ReLU (Rectified Linear Unit) activation functions instead of tanh (hyperbolic tangent), whichever helped minimize the network’s training time and was a present solution to the ”vanishing gradient” difficulty at the time. When constructing the feature map, the pooling layers
ISO 9001:2008 Certified Journal
|
Page 295