Unsupervised Trademark Retrieval Method Based on Attention Mechanism

Aiming at the high cost of data labeling and ignoring the internal relevance of features in existing trademark retrieval methods, this paper proposes an unsupervised trademark retrieval method based on attention mechanism. In the proposed method, the instance discrimination framework is adopted and a lightweight attention mechanism is introduced to allocate a more reasonable learning weight to key features. With an unsupervised way, this proposed method can obtain good feature representation of trademarks and improve the performance of trademark retrieval. Extensive comparative experiments on the METU trademark dataset are conducted. The experimental results show that the proposed method is significantly better than traditional trademark retrieval methods and most existing supervised learning methods. The proposed method obtained a smaller value of NAR (Normalized Average Rank) at 0.051, which verifies the effectiveness of the proposed method in trademark retrieval.

Implementation of optical flow: good feature definition for tracking of heart cavity

<span>Echocardiography is a method of examination using high-frequency sound waves to capture images of the heart organ structure. Echocardiography video is used by a doctor to analyze heart wall cavity movements and identify heart disease. Several points of view including the long axis, the two and four cavities in the left ventricle can be used in the examination of heart function. Cardiac assessment is still performed conventionally, which requires a level of thoroughness. This research proposes a method for tracking the movement of the heart wall. In this method, the good feature was defined only in the first frame. Furthermore, the whole frame will be processed by the optical flow method. Good feature definition consists of image enhancement, segmentation and tracking processes using the optical flow. Furthermore, the calculation of contour similarity from the proposed method by forming contour manually using 24 point initialization to draw the heart cavity. The proposed method shows the calculation results with a sensitivity of 90% and an accuracy of 87.451%.</span>

Remote Sensing-Guided Sampling Design with Both Good Spatial Coverage and Feature Space Coverage for Accurate Farm Field-Level Soil Mapping

With the increasing requirements of precision agriculture for massive and various kinds of data, remote sensing technology has become indispensable in acquiring the necessary data for precision agriculture. Understanding the spatial variability of a target soil variable (i.e., soil mapping) is a critical issue in solving many agricultural problems. Field sampling is one of the most commonly used technologies for soil mapping, but sample sizes are restricted by resources, such as field labor, soil physicochemical analysis, and funding. In this paper, we proposed a sampling design method with both good spatial coverage and feature space coverage to achieve more precise spatial variability of farm field-level target soil variables for limited sample sizes. The proposed method used the super-grid to achieve good spatial coverage, and it took advantage of remote sensing products that were highly correlated with the target soil property (SOM content) to achieve good feature space coverage. For the experiments, we employed the ordinary kriging (OK) method to map the soil organic matter (SOM) content. The different sized super-grid comparison experiments showed that the 400 × 400 m2 super-grid had the highest SOM content mapping accuracy. Then, we compared the proposed method to regular grid sampling (good spatial coverage) and k-means sampling (good feature space coverage), and the experimental results indicated that the proposed method had greater potential in the selection of representative samples that could improve the SOM content mapping accuracy.

A new invariant of 4-manifolds

We define an integer invariant LX of a smooth, compact, closed 4-manifold X by minimizing a certain complexity of a trisection of X over all trisections. The good feature of LX is that when LX=0 and X is a homology 4-sphere, then X is diffeomorphic to the 4-sphere. Naturally, L is hard to compute.

RESEARCH ON METHOD OF INTERACTIVE SEGMENTATION BASED ON REMOTE SENSING IMAGES

In this paper, we aim to solve the object extraction problem in remote sensing images using interactive segmentation tools. Firstly, an overview of the interactive segmentation algorithm is proposed. Then, our detailed implementation of intelligent scissors and GrabCut for remote sensing images is described. Finally, several experiments on different typical features (water area, vegetation) in remote sensing images are performed respectively. Compared with the manual result, it indicates that our tools maintain good feature boundaries and show good performance.