Object–Part Registration–Fusion Net for Fine-Grained Image Classification

Classifying fine-grained categories (e.g., bird species, car, and aircraft types) is a crucial problem in image understanding and is difficult due to intra-class and inter-class variance. Most of the existing fine-grained approaches individually utilize various parts and local information of objects to improve the classification accuracy but neglect the mechanism of the feature fusion between the object (global) and object’s parts (local) to reinforce fine-grained features. In this paper, we present a novel framework, namely object–part registration–fusion Net (OR-Net), which considers the mechanism of registration and fusion between an object (global) and its parts’ (local) features for fine-grained classification. Our model learns the fine-grained features from the object of global and local regions and fuses these features with the registration mechanism to reinforce each region’s characteristics in the feature maps. Precisely, OR-Net consists of: (1) a multi-stream feature extraction net, which generates features with global and various local regions of objects; (2) a registration–fusion feature module calculates the dimension and location relationships between global (object) regions and local (parts) regions to generate the registration information and fuses the local features into the global features with registration information to generate the fine-grained feature. Experiments execute symmetric GPU devices with symmetric mini-batch to verify that OR-Net surpasses the state-of-the-art approaches on CUB-200-2011 (Birds), Stanford-Cars, and Stanford-Aircraft datasets.

Interpretable Compositional Convolutional Neural Networks

This paper proposes a method to modify a traditional convolutional neural network (CNN) into an interpretable compositional CNN, in order to learn filters that encode meaningful visual patterns in intermediate convolutional layers. In a compositional CNN, each filter is supposed to consistently represent a specific compositional object part or image region with a clear meaning. The compositional CNN learns from image labels for classification without any annotations of parts or regions for supervision. Our method can be broadly applied to different types of CNNs. Experiments have demonstrated the effectiveness of our method. The code will be released when the paper is accepted.

Fine-grained classification based on multi-scale pyramid convolution networks

The large intra-class variance and small inter-class variance are the key factor affecting fine-grained image classification. Recently, some algorithms have been more accurate and efficient. However, these methods ignore the multi-scale information of the network, resulting in insufficient ability to capture subtle changes. To solve this problem, a weakly supervised fine-grained classification network based on multi-scale pyramid is proposed in this paper. It uses pyramid convolution kernel to replace ordinary convolution kernel in residual network, which can expand the receptive field of the convolution kernel and use complementary information of different scales. Meanwhile, the weakly supervised data augmentation network (WS-DAN) is used to prevent over fitting and improve the performance of the model. In addition, a new attention module, which includes spatial attention and channel attention, is introduced to pay more attention to the object part in the image. The comprehensive experiments are carried out on three public benchmarks. It shows that the proposed method can extract subtle feature and achieve classification effectively.

On the Use of Standardized Multi-Temporal Indices for Monitoring Disturbance and Ecosystem Moisture Stress across Multiple Earth Observation Systems in the Google Earth Engine

In this work we explore three methods for quantifying ecosystem vegetation responses spatially and temporally using Google’s Earth Engine, implementing an Ecosystem Moisture Stress Index (EMSI) to monitor vegetation health in agricultural, pastoral, and natural landscapes across the entire era of spaceborne remote sensing. EMSI is the multitemporal standard (z) score of the Normalized Difference Vegetation Index (NDVI) given as I, for a pixel (x,y) at the observational period t. The EMSI is calculated as: zxyt = (Ixyt − ?xyT)/?xyT, where the index value of the observational date (Ixyt) is subtracted from the mean (?xyT) of the same date or range of days in a reference time series of length T (in years), divided by the standard deviation (?xyT), during the same day or range of dates in the reference time series. EMSI exhibits high significance (z > |2.0 ± 1.98σ|) across all geographic locations and time periods examined. Our results provide an expanded basis for detection and monitoring: (i) ecosystem phenology and health; (ii) wildfire potential or burn severity; (iii) herbivory; (iv) changes in ecosystem resilience; and (v) change and intensity of land use practices. We provide the code and analysis tools as a research object, part of the findable, accessible, interoperable, reusable (FAIR) data principles.

Identification of Freshness of Marine Fish Based on Image of Hue Saturation Value and Morphology

Euthynus is one of the fish that is widely consumed for the enjoyment of the people of Indonesia or abroad, because of its very soft quality, easy to obtain, and contains a lot of essential protein amino acids that are good for the body. This research aims to identify the freshness of the fish purchased based on the eyes and fish gills. The initial process of identifying the freshness of fish uses several methods. Image input process through image object taking using a cell phone camera. The image object is used to determine the value of the RGB image object. RGB color extraction clarifies the value obtained from the image object before proceeding to the next process. Image resize is the process of cutting the image on the desired object part. Image conversion using the HSV method was used to determine the freshness of fish in the gills. The Local Binary Pattern method is used to determine the freshness of the fisheye. The next step is to refine the RGB image into Morphology. The KNN (K-Nearest Neighbor Method) method is used to group objects based on learning data closest to the object. The journal analysis results on the comparison of methods, after 45 trials for each method, found that the Hue Saturation Value method obtained the highest success by 90% and for the texture value obtained 85% success.

Mihi est from Brythonic to Breton I

Middle Breton (MB) presents a singular anomaly of pronominal argument coding. Objects are accusative proclitics save in two constructions, where coding is split by person: 3rd unique enclitics ~ 1st/2nd accusative proclitics. The constructions are HAVE, from Insular Celtic mihi est, where the new coding replaces inflectional nominatives (cf. Latin mihi est ~ sunt); and imperatives, where it replaces accusative enclitics in V1 (cf. French aide-moi ~ ne m’aide pas). The evolution is traced in light of a crosslinguistic construction type that suggests its nature, noncanonical subject + 3rd nominative ~ 1st/2nd accusative object. Part I: (1) Decomposition of HAVE as dative clitic + BE from Brythonic throughout “conservative” varieties of Breton. (2) Breton-Cornish innovation of nonclitic datives for mihi est and their subjecthood. Part II: (3) Brythonic unavailibility of mesoclisis in V1 and Breton-Cornish nonagreement with nominative objects, resulting in independent > enclitic pronouns for accusative objects of imperatives and nominative objects of mihi est. (4) MB alignment of imperatives with mihi est in 3rd person, restriction on nominative enclitics, and recruitment of 1st/2nd person accusative proclitics upon loss of mesoclisis. (5) Transition to accusative objects in “innovative” varieties and subject-object case interactions.

Magnetostatic Active Contour Model with Classification Method of Sparse Representation

The active contour model is widely used to segment images. For the classical magnetostatic active contour (MAC) model, the magnetic field is computed based on the detected points by using an edge detector. However, noise and nontarget points are always detected. Thus, MAC is nonrobust to noise and the extracted objects may be deviant from the real objects. In this paper, a magnetostatic active contour model with a classification method of sparse representation is proposed. First, rough edge information is obtained with some edge detectors. Second, the extracted edge contours are divided into two parts by sparse classification, that is, the target object part and the redundant part. Based on the classified target points, a new magnetic field is generated, and contours evolve with MAC to extract the target objects. Experimental results show that the proposed model could decrease the influence of noise and robust segmentation results could be obtained.