Clinical keratoconus (KCN) detection is a challenging and time-consuming task. In the diagnosis process, ophthalmologists must revise demographic and clinical ophthalmic examinations. The latter include slit-lamb, corneal topographic maps, and Pentacam indices (PI). We propose an Ensemble of Deep Transfer Learning (EDTL) based on corneal topographic maps. We consider four pretrained networks, SqueezeNet (SqN), AlexNet (AN), ShuffleNet (SfN), and MobileNet-v2 (MN), and fine-tune them on a dataset of KCN and normal cases, each including four topographic maps. We also consider a PI classifier. Then, our EDTL method combines the output probabilities of each of the five classifiers to obtain a decision based on the fusion of probabilities. Individually, the classifier based on PI achieved 93.1% accuracy, whereas the deep classifiers reached classification accuracies over 90% only in isolated cases. Overall, the average accuracy of the deep networks over the four corneal maps ranged from 86% (SfN) to 89.9% (AN). The classifier ensemble increased the accuracy of the deep classifiers based on corneal maps to values ranging (92.2% to 93.1%) for SqN and (93.1% to 94.8%) for AN. Including in the ensemble-specific combinations of corneal maps’ classifiers and PI increased the accuracy to 98.3%. Moreover, visualization of first learner filters in the networks and Grad-CAMs confirmed that the networks had learned relevant clinical features. This study shows the potential of creating ensembles of deep classifiers fine-tuned with a transfer learning strategy as it resulted in an improved accuracy while showing learnable filters and Grad-CAMs that agree with clinical knowledge. This is a step further towards the potential clinical deployment of an improved computer-assisted diagnosis system for KCN detection to help ophthalmologists to confirm the clinical decision and to perform fast and accurate KCN treatment.
One of the troublesome duties in chemical industrial units is determining the instantaneous drop size distribution, which is created between two immiscible liquids within such units. In this work a complete system for measuring instantaneous droplet size is constructed. It consists of laser detection system (1mW He-Ne laser), drop generation system (turbine mixer unit), and microphotography system. Two immiscible liquids, water and kerosene were mixed together with different low volume fractions (0.0025, 0.02) of kerosene (as a dispersed phase) in water (as a continuous phase). The experiments were carried out at different rotational speed (1180- 2090 r.p.m) of the turbine mixer. The Sauter mean diameter of the drops was determined by la
... Show MoreIn this work, satellite images classification for Al Chabaish marshes and the area surrounding district in (Dhi Qar) province for years 1990,2000 and 2015 using two software programming (MATLAB 7.11 and ERDAS imagine 2014) is presented. Proposed supervised classification method (Modified Vector Quantization) using MATLAB software and supervised classification method (Maximum likelihood Classifier) using ERDAS imagine have been used, in order to get most accurate results and compare these methods. The changes that taken place in year 2000 comparing with 1990 and in year 2015 comparing with 2000 are calculated. The results from classification indicated that water and vegetation are decreased, while barren land, alluvial soil and shallow water
... Show MoreA number of compression schemes were put forward to achieve high compression factors with high image quality at a low computational time. In this paper, a combined transform coding scheme is proposed which is based on discrete wavelet (DWT) and discrete cosine (DCT) transforms with an added new enhancement method, which is the sliding run length encoding (SRLE) technique, to further improve compression. The advantages of the wavelet and the discrete cosine transforms were utilized to encode the image. This first step involves transforming the color components of the image from RGB to YUV planes to acquire the advantage of the existing spectral correlation and consequently gaining more compression. DWT is then applied to the Y, U and V col
... Show MoreA low-cost, RGB LED-based visible-light spectrophotometer was designed to measure dyes concentration. Dyes are widely used as indicators or coloring agents in different applications and knowing their concentration is an essential part for many studies. The proposed spectrophotometer provides many functionalities that clones the traditional expensive spectrophotometers for a budged price under $50. It was aimed to provide a versatile tool for instructors and educators to teach their students the fundamental concepts behind spectrophotometry. Malachite green, methyl red, and methyl orange dyes were chosen to be good samples to show the integrity of the proposed spectrophotometer in terms of accuracy, repeatability, and sensitivity as
... Show MoreThe type of video that used in this proposed hiding a secret information technique is .AVI; the proposed technique of a data hiding to embed a secret information into video frames by using Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT) and Curvelet Transform (CvT). An individual pixel consists of three color components (RGB), the secret information is embedded in Red (R) color channel. On the receiver side, the secret information is extracted from received video. After extracting secret information, robustness of proposed hiding a secret information technique is measured and obtained by computing the degradation of the extracted secret information by comparing it with the original secret information via calculating the No
... Show MoreThe art of preventing the detection of hidden information messages is the way that steganography work. Several algorithms have been proposed for steganographic techniques. A major portion of these algorithms is specified for image steganography because the image has a high level of redundancy. This paper proposed an image steganography technique using a dynamic threshold produced by the discrete cosine coefficient. After dividing the green and blue channel of the cover image into 1*3-pixel blocks, check if any bits of green channel block less or equal to threshold then start to store the secret bits in blue channel block, and to increase the security not all bits in the chosen block used to store the secret bits. Firstly, store in the cente
... Show MoreThe microbend sensor is designed to experience a light loss when force is applied to the sensor. The periodic microbends cause propagating light to couple into higher order modes, the existing higher order modes become unguided modes. Three models of deform cells are fabricated at (3, 5, 8) mm pitchand tested by using MMF and laser source at 850 nm. The maximum output power of (8, 5, 3)mm model is (3, 2.7, 2.55)nW respectively at applied force 5N and the minimum value is (1.9, 1.65, 1.5)nW respectively at 60N.The strain is calculated at different microbend cells ,and the best sensitivity of this sensor for cell 8mm is equal to 0.6nW/N.