Texture classification using convolutional neural network optimized with whale optimization algorithm

Research Article Texture classification using convolutional neural network optimized with whale optimization algorithm Ujjawal Dixit1 · Apoorva Mishra2 · Anupam Shukla3 · Ritu Tiwari1 © Springer Nature Switzerland AG 2019 Abstract Texture classification is an active area of research in the field of pattern recognition. Convolutional neural networks (CNNs) have a remarkable capability of recognizing patterns and are one of the most efficient deep learning techniques. But, finding the optimal values of the different hyperparameters of the CNN is a major challenge. Nature-inspired algorithms (NIAs) are the meta-heuristic algorithms well-known for their optimizing capability. Whale optimization algorithm (WOA) is a recent nature-inspired algorithm (NIA) that is inspired by the hunting behaviour of the humpback whales. In this paper, we propose a novel deep learning technique for texture recognition using a CNN optimized through WOA. We apply WOA at the two different levels in the CNN: In the convolutional layer (for optimizing the values of the filters), and in the fully-connected layer (for optimizing the values of the weights and biases). For examining the performance of our technique, we apply it to the following three benchmark texture datasets: Kylberg v1.0, Brodatz, and Outex_TC_00012. Our model performs better than the most of the existing methods for the Kylberg and the Outex_TC_00012 datasets and gives competitive results for the Brodatz dataset. It is evident from the results that our model has the potential for application in the field of texture recognition. Keywords Convolutional neural network · Whale optimization algorithm · Pattern recognition · Texture classification · Deep learning 1 Introduction 1.1 CNN Deep Learning for the past few years has evolved as one of the important pillars while developing models based on machine learning. CNN is one of the models which have performed exceptionally well for image classification and pattern recognition tasks [1, 2]. The following sub-sections help in introducing the fundamentals of texture recognition, CNN, and NIA and also discuss the contributions of the paper and its organization. CNN is a biologically inspired technique for classification. It generally deals with the image classification and pattern recognition tasks. The architecture of a simple CNN is represented by Fig. 1 as shown below. The working of a typical CNN is explained as follows. The data in the input layer is being processed by the convolutional layer with the help of filters/kernels to generate feature maps which signify the raw features. The pooling layer performs a downsampling operation which reduces the dimensionality of the feature map. The processed features from the ReLU units are supplied to the fully connected layer (FCL) which enables classification of the data. * Apoorva Mishra, | 1Soft Computing and Expert System Laboratory, ABV-IIITM, Gwalior, Gwalior, India. 2Department of Computer Science Engineering, Bennett University, Greater Noida, India. 3Indian Institute of Information Technology, Pune, Pune, India. SN Applied Sciences (2019) 1:655 | https://doi.org/10.1007/s42452-019-0678-y Received: 1 January 2019 / Accepted: 28 May 2019 / Published online: 31 May 2019 Vol.:(0123456789) Research Article SN Applied Sciences (2019) 1:655 | https://doi.org/10.1007/s42452-019-0678-y and “spots.” The convolution operation plays a crucial role in determining the neighbouring properties. Hence, the techniques which are based on the convolutional operation can be helpful in the classification of the textures. CNN is one such technique which helps in identifying the local neighbourhood properties during the feature extraction phase. In this paper, we have utilized this property of the CNN to excel over the texture databases. 1.3 Nature inspired algorithm Fig. 1  Basic CNN model Output layer generally consists of the soft-max approximation which facilitates the multiclass classification (if needed). 1.2 Texture classification Texture is the atomic quantity for the characterization of an object which helps in its identification. Various images such as medical, agricultural, aerial, satellite and others have been identifiable due to the presence of texture in them. Hence, textures play a key role in distinguishing objects in such images. In the textile industry as well, textures play an important role. Hence, the correct classification of texture could be very useful in such fields and forms the motivation for conducting the current research. In the recent years, textures have been widely used in the content-based image retrieval systems as well. The common methods which are used for texture classification are namely parametric statistical model-based methods, structural methods, empirical second order statistical methods and various other transform methods. Deep learningbased techniques for the classification of texture have been proposed in [3–6]. There is also a major class of classification approaches based on the local properties of texture. These approaches use local operators to identify the local properties of texture and classify them according to those properties. Also, the local neighbourhood properties are the dominant reasons for the overall appearance of a texture or a pattern, and so, any local operation on exploiting the neighbourhood properties helps in identifying the texture. The common local properties which are exploited are “edges” Vol:.(1234567890) NIAs are the meta-heuristic algorithms which have the remarkable capability to solve optimization problems concerning the constrained environment. Most of these problems are NP-hard in nature and cannot be solved using the traditional deterministic algorithms. NIAs have been proven to be an excellent method to address these complex optimization problems, and have been applied to solve many such problems. Over the past few decades, various NIAs have been developed taking inspiration from the processes that occur in nature; WOA is one such recently developed algorithm [7]. Recently, NIAs have been applied to the fields of medical image classification, robot path planning, financial and industrial optimization, etc. through hybridization with various existing machine learning techniques [8]. 1.3.1 Whale optimization algorithm WOA is an NIA that imitates the behaviour of humpback whales [7]. WOA has been hybridized with the various machine learning algorithms like SVM, ANN, etc. [9–12]. WOA consists of the following two phases [7]. I. II. Encircling Prey (Exploration Phase) and, Bubble-Net Attacking (Exploitation Phase) The basic WOA can be mathematically represented as follows [7]. ⃗ = 2a ���⃗ ⋅ ⃗r − a⃗ A (1) C⃗ = 2 ⋅ ⃗r (2) ⃗ = ||C⃗ ⋅ X⃗ ∗ (t) − X(t) ⃗ || D | | (3) ⃗ ⃗ + 1) = X⃗ ∗ (t) − A ⃗⋅D X(t (4) ⃗ = ||C⃗ ⋅ X⃗rand − X⃗ || D | | (5) ⃗ ⃗ + 1) = X⃗rand − A ⃗⋅D X(t (6) SN Applied Sciences (2019) 1:655 | https://doi.org/10.1007/s42452-019-0678-y (...truncated)