Reference links:
https://blog.csdn.net/hongxingabc/article/details/78833485
aerial remote sensing image has a particularity:
1 scale diversity: aerial remote sensing images taken from a few hundred meters to nearly 10,000 meters of height are, and ground targets even sizes similar goals, such as a large harbor vessels with more than 300 meters, and only a small few tens of meters;
2. the particularity of perspective: the perspective of aerial remote sensing images are basically high top, but regular data most of the ground set or horizontal viewing angle, so the same object model is different, in the conventional data set training good detector, the possible effect on the poor aerial remote sensing image;
3. small target issues: remote sensing image aviation the goal of many of which are small targets (dozens or even a few pixels), which leads to less objective information, CNN target detection method based on a conventional target detection Yiqijuechen data set, but for the small target, CNN Pooling the information layer will be further reduced, the target of a 24 * 24 after 4 pooling layer only about one pixel, so that the low dimension indistinguishable. It does not seem related to the visibility of paper before, should be able to draw small objects mentioned in the previous paper.
4. The use of a top aerial remote sensing image capturing, the direction of the target is uncertain (and often have some certainty conventional data set, such as a pedestrian, the vehicle basically standing), the target detector needs to have a direction of Lu stick of.
Background The high complexity, aviation remote sensing image field of view is relatively large (typically several kilometers of coverage), the field of view may contain a variety of contexts, the detection target will produce a strong interference.
Thus, there are several data sets:
DOTA, the UCAS-the AOD, NWPU VHR-10, a Dataset-RSOD, INRIA Aerial Image DataSet
Download the NPU data sets: a total of 10 categories. Related papers are as follows:
- G. Cheng, J. Han, P. Zhou, L. Guo. Multi-class geospatial object detection and geographic imageclassification based on collection of part detectors. ISPRS Journal ofPhotogrammetry and Remote Sensing, 98: 119-132, 2014.
- You need to register, as if unable to receive mail
SCI journal Remote Sensing:
https://blog.csdn.net/dovejay/article/details/78230613: name of the magazine and contributor experience
https://blog.csdn.net/qq10593994/article/details/50992599
like tgrs (IEEE TRANSACTIONS ON GEOSCIENCE aND REMOTE SENSING) and isprs (ISPRS Journal of Photogrammetry andRemote Sensing ) are a top field of remote sensing published.
If there is something more innovative, we have preferred tgrs. rse and tgrs mainstream journals of remote sensing image processing / imaging model. isprs before adding weng editor, focusing on photogrammetry and lidar direction. Compared environmental remote sensing and remote sensing images, this direction is really a very small minority, so little influence is understandable.
For data, in the year 2000, isprs published 23 papers, rse a total of 128, tgrs a total of 257, so the history of how the influence of these three could be as?
So isprs been criticized in the historical heritage is normal (of course, in the direction of photogrammetry and lidar, this journal has been the mainstream).
After adding weng, ecological remote sensing, image algorithms, and even papers spatial information processing more and more together.
In recent years it gradually transformed from a small journals direction for a like tgrs / rse as large remote sensing journals.
It's progress for all to see, after a large influence in the RS will gradually accumulation. In addition, as a next isprs Journal commented that, if the algorithm is really amazing,
completely out of the three so-called good journals journals Mapping and Remote Sensing, such as excellent paper image processing / lidar, and published in IJCV, IEEE TPAMI , IEEE TIP often have on the (top journals).
Three journals: rse, tgrs, isprs
IEEE TPAMI and IJCV is the top international journal of computer vision and artificial intelligence, CVPR, ICCV and ECCV are the top three international conferences in computer vision, NIPS is the top international conferences artificial intelligence, which the IEEE TPAMI, IJCV and IEEE TIP is recommended CCF A class international journals, CVPR, ICCV and CCF recommended NIPS is a class A international conference (see China computer Federation (CCF) recommended by the international Conference Journal list: http://www.ccf.org.cn/xspj/gyml/ ).
We found several articles on object detection field of remote sensing images from the meeting of the CCF, and download down.