Using Cross Entropy as a Performance Metric for Quantifying Uncertainty in DNN Image Classifiers: An Application to Classification of Lung Cancer on CT Images

Cross entropy is a measure in machine learning and deep learning that assesses the difference between predicted and actual probability distributions. In this study, we propose cross entropy as a performance evaluation metric for image classifier models and apply it to the CT image classification of lung cancer. A convolutional neural network is employed as the deep neural network (DNN) image classifier, with the residual network (ResNet) 50 chosen as the DNN archi-tecture. The image data used comprise a lung CT image set. Two classification models are built from datasets with varying amounts of data, and lung cancer is categorized into four classes using 10-fold cross-validation. Furthermore, we employ t-distributed stochastic neighbor embedding to visually explain the data distribution after classification. Experimental results demonstrate that cross en-tropy is a highly useful metric for evaluating the reliability of image classifier models. It is noted that for a more comprehensive evaluation of model perfor-mance, combining with other evaluation metrics is considered essential. .

Defend Against Adversarial Samples by Using Perceptual Hash

Image classifiers that based on Deep Neural Networks(DNNs)have been proved to be easily fooled by well-designed perturbations.Previous defense methods have the limitations of requiring expensive computation or reducing the accuracy of the image classifiers.In this paper,we propose a novel defense method which based on perceptual hash.Our main goal is to destroy the process of perturbations generation by comparing the similarities of images thus achieve the purpose of defense.To verify our idea,we defended against two main attack methods(a white-box attack and a black-box attack)in different DNN-based image classifiers and show that,after using our defense method,the attack-success-rate for all DNN-based image classifiers decreases significantly.More specifically,for the white-box attack,the attack-success-rate is reduced by an average of 36.3%.For the black-box attack,the average attack-success-rate of targeted attack and non-targeted attack has been reduced by 72.8%and 76.7%respectively.The proposed method is a simple and effective defense method and provides a new way to defend against adversarial samples.

Mapping Deciduous Broad-leaved Forested Swamps Using ALOS/Palsar Data

Accurate, updated information on the distribution of wetlands is essential for estimating net fluxes of greenhouse gases and for effectively protecting and managing wetlands. Because of their complex community structure and rich surface vegetation, deciduous broad-leaved forested swamps are considered to be one of the most difficult types of wetland to classify. In this research, with the support of remote sensing and geographic information system, multi-temporal radar images L-Palsar were used initially to extract the forest hydrological layer and phenology phase change layer as two variables through image analysis. Second, based on the environmental characteristics of forested swamps, three decision tree classifiers derived from the two variables were constructed to explore effective methods to identify deciduous broad-leaved forested swamps. Third, this study focused on analyzing the classification process between flat-forests, which are the most severely disturbed elements, and forested swamps. Finally, the application of the decision tree model will be discussed. The results showed that: 1) L-HH band(a L band with wavelength of 0–235 m in HH polarization mode; HH means Synthetic Aperture Radars transmit pulses in horizontal polarization and receive in horizontal polarization) in the leaf-off season is shown to be capable of detecting hydrologic conditions beneath the forest; 2) the accuracy of the classification(forested swamp and forest plat) was 81.5% based on hydrologic features, and 83.5% was achieved by combining hydrologic features and phenology response features, which indicated that hydrological characteristics under the forest played a key role. The HHOJ(refers to the band created by the subtraction with HH band in October and HH band in July) achieved by multi-temporal radar images did improve the classification accuracy, but not significantly, and more leaf-off radar images may be more efficient than multi-seasonal radar images for inland forested swamp mapping; 3) the lower separability between forested swamps dominated by vegetated surfaces and forest plat covered with litter was the main cause of the uncertainty in classification, which led to misleading interpretations of the pixel-based classification. Finally, through the analysis with kappa coefficients, it was shown that the value of the intersection point was an ideal choice for the variable.

Foliar fungal disease classification in banana plants using elliptical local binary pattern on multiresolution dual tree complex wavelet transform domain

The fungal diseases in banana cause major yield losses for millions of farmers around the globe.Early detection of these diseases helps the farmers to devise successful management strategies.The characteristic leaf blade discoloration pattern at the earlier stages of infection could be used to understand the onset of each disease.This paper demonstrates a methodology for classification of three important foliar diseases in banana,using local texture features.The disease affected regions are identified using image enhancement and color segmentation.Segmented images are converted to transform domain using three image transforms(DWT,DTCWT and Ranklet transform).Feature vector is extracted from transform domain images using LBP and its variants(ELBP,MeanELBP and MedianELBP).These texture based features are applied to five popular image classifiers and comparative performance analysis is done using ten-fold cross validation procedure.Experimental results showed best classification performance for ELBP features extracted from DTCWT domain(accuracy 95.4%,precision 93.2%,sensitivity 93.0%,Fscore 93.0%and specificity 96.4%).Compared with traditional methods of feature extraction,this novel method of fusing DTCWT with ELBP features has attained high degree of accuracy in precisely detecting and classifying fungal diseases in banana at an early stage.