ABSTRACT In recent years, deep learning algorithm has been used in many applications mainly in image processing of object detection and classification. The use of image processing techniques is becoming more interesting with the existence of drone technology with the employ of deep learning in aerial view image processing because of the high resolution and heaps of images taken. This paper aims to review neural networks specifically on the aerial view image by drones and to discuss the work principles and classic architectures of convolutional neural networks, its latest research trend and typical models along with target detection in object detection, image classification and semantic segmentation. In addition, this study also provided a specific application in the aerial image. Finally, the limitations of the convolutional network and expected future development trends were also discussed. Based on the findings, the deep learning algorithm was observed to provide high accuracy, it outperformed other generally image processing-based techniques.

ABSTRACT Uses of video surveillance (VS) have exponentially increased using the internet as a platform. Security concerns related to the transmission of videos must be addressed. Even though the majority of current encryption standards are effective, it is computationally demanding and unsuitable for real-time VS applications. Lightweight crypto-encryption standards are necessary to meet real-time needs. This paper provides a thorough analysis of various encryption standards to protect video data and evaluates vulnerability under various attacks. A Lightweight Fast Security Standard (LFSS) is proposed to meet real-time requirements by adopting dual key fusion (KF) approach. Entropy and crypto weights are quantitatively compared with proposed and chosen lightweight ciphers for surveillance videos based on NCPR, UACI and attacks into account. The proposed LFSS method offers a good NCPR of 99.5806 and offers only 40% and 3% average execution time over Secure Internet of things (SIT) and Secure Force (SF) security standards, respectively.

ABSTRACT The facial expressions recognition (FER) is crucial to many applications. As technology advances and our needs evolve, compound emotion recognition is becoming increasingly important, along with basic emotion recognition. In the literature, Although, FER can be conducted primarily using multiple sensors. However, research shows that using facial images/videos to recognize facial expressions is better because visual presentation can convey more efficiently. Among state-of-the-art methods for FER systems, to improve the accuracy of the basic and compound FER systems, detection of facial action units (AUs) must be combined to detect basic and compound facial expressions. State-of-the-art results show that machine learning and deep learning-based approaches are more potent than conventional FER approaches. This paper surveys various learning frameworks for facial emotion recognition systems for detecting basic and compound emotions using the diverse database and summarizing state-of-the-art results to give good understanding of impact of each learning framework used in FER systems.

ABSTRACT Plant disease phenotyping using technology assures a promising step in sustainable agriculture. Solanum Tubrosum (Potato) is a highly cultivated vegetable plant, which is affected by fungal pathogens that leads to Early Blight and Late Blight diseases. Continuous monitoring of plant disease is a challenging task in identifying diseased leaves. This paper proposes a novel deep Convolutional Neural Network (CNN) for classifying the potato disease by reducing the computational time of learnable parameters using image phenotyping. A meta-heuristic algorithm known as the Whale Optimization Algorithm (WOA) is used in training to optimize the hyperparameters of the proposed CNN network. Potato disease is classified using optimized CNN called POT-Net, and the performance is analyzed using performance metrics: precision, recall, F1-score, and accuracy of each class. The performance POT-Net is compared with pre-trained DL and optimized algorithms using performance metrics and it is better than the state-of-the-art models, with an accuracy of 99.12%.

ABSTRACT Ventricle pathology changes and their severity overlap among the dementia classes to understand the pathogenesis of this disorder. In the present work, left and right ventricle variations in the severity classes of dementia were observed using optimizations, along with the dual deep learning techniques (DDLT). Segmentation of ventricle region was carried out using a multilevel threshold-based grey wolf optimization and palindrome detection method was executed to identify its symmetry. AlexNet and ResNet were used to extract the DDLT features which are then used for classification. The obtained results showed that the ventricle region was accurately delineated with a higher degree of correspondence which was >0.9. Furthermore, it was observed that the DDLT with multi-class SVM provided improved accuracy in the left ventricle with 84.8% and the right ventricle with 81.2%. Thus, the left ventricle variation was claimed to be a distinct indicator in demarcating different classes of dementia.

ABSTRACT This work presents an object recognition algorithm that combines local binary pattern (LBP) based on fuzzy logic approach and active contour model for segmenting different images to detect textured objects. Initially, images containing objects are segmented using the fuzzy logic-optimized LBP method. Then, we eliminate the image noise. Finally, utilizing a Chan-Vese active contour method, the target object of the image is highlighted. The segmentation has been compared with the classical LBP technique and the results indicated higher accuracy and quality for highlighting the object from the background. The classification of highlighted objects is performed with a convolution neural network (CNN). To authenticate the proposed approach, 140 images with the classification of 10 different objects were used. The simulation depicted that the proposed method has better results than other methods both in terms of segmentation error and performance. Significantly, CNN classification also showed a classification accuracy of 92.8%.

ABSTRACT Among the various cancer categories, breast cancer diagnoses are commonly found in many women, and automatic classification of breast cancer images is a crucial task using computer-aided analysis. This paper proposed a method for the detection of histopathological image-based breast cancer and the main motive of the technique is to employ feature extraction and classification based on DL techniques. The input image is filtered using edge smoothening for noise removal and image resizing on the pre-processing method and used 3D-Convolutional Neural network (3D_CNN) for extracting the features of the image process. Extracted deep features have been given as input for the classification process using Stochastic Diffusion Kernel Recursive Neural Networks (SDKRNN). The suggested model combines the sturdiness of convolutional and kernel blocks and demonstrates performance stability when compared to existing techniques. The experimental results are assessed using a variety of performance metrics and compared to current breast cancer detection methodologies. The result demonstrated the experimental evaluation analyzed for various datasets in terms of accuracy, precision, F-1 score, specificity, and AUC of 98%, 93.8%, 89%, 80%, and 70% and it determined the outperforms in the detection of breast cancer.

ABSTRACT Ship classification has long been a challenging task in synthetic aperture radar (SAR) applications. With the improvement of SAR resolution, it has gradually become possible based on high-resolution SAR images. This letter presents a joint feature representation method for characterizing the superstructure scattering features of merchant ships, and then a hierarchical scheme for classifying merchant ships in high-resolution SAR images based on integrated macro and detailed superstructure features is proposed. Comprehensive experiment on three types of high-resolution SAR samples demonstrates that the proposed features and method have the ability to improve the performance of merchant ship classification in SAR images.

ABSTRACT This paper develops an effective encryption and steganography-based text extraction in IoT using deep learning method. Initially, the input text and cover images are separately pre-processed. DCT (discrete cosine transform) is utilized to transfer the image from spatial domain to frequency domain. Then, the original text is encrypted using new optimized equilibrium-based homomorphic encryption (OEHE) approach. Next, the extended wavelet convolutional transient search (EWCTS) optimizer with quotient multi-pixel value differencing (QMPVD) is developed to embed the secret text in cover images. Then, at receiver side, the reverse process for encryption and steganography is executed with secret key provided by the sender. Finally, the accurate text is extracted at receiver side using steganalysis process. The developed approach is executed in MATLAB software. The various evaluation metrics are used to authorize the effectiveness of suggested approach. Simulation outcomes proved that the suggested technique provides better outcomes than other existing approaches.

ABSTRACT The non-invasive, cost-effective and radiation-free methodology of ultrasound imaging procedures makes it a convenient tool for the detection of tumours. But, usually, the diagnosis efficiency and accuracy of detection is affected by speckle, intensity inhomogeneities and low contrast. For tumour detection, identifying and segmenting the tumour region is vital. In this proposed work, active contours based segmentation is employed with the initial contour achieved using a thresholding method involving a novel concept of phase gradients. Initially, the raw US image is de-speckled and enhanced using the neutrosophic transformation concept. This pre-processed US image is then thresholded using adaptive constraints derived from the mean, texture (using Gabor filters), gradient, phase information and phase gradient of the image. Then the obtained thresholded image is used as the initial mask/contour for segmentation based on active contours using localized image energies. The proposed methodology gave improved tumour boundary delineations even for challenging tumour echogenicities.