Search

The Landsat program has a long history of providing remotely sensed data to the user community. This history is being extended with the addition of the Landsat 9 satellite, which closely mimics the Landsat 8 satellite and its instruments. These satellites contain two instruments, the Operational Land Imager (OLI) and the Thermal Infrared Sensor (TIRS). OLI is a push-broom sensor that collects visible and near-infrared (VNIR) and short-wave infrared (SWIR) wavelengths at 30 m ground sample distance, along with a panchromatic 15 m band. The TIRS sensor contains two long-wave thermal spectral channels centered at 10.9 and 12 µm. The data from these two instruments, on both satellites, are combined into a single Landsat product. The Landsat 5–9 satellites follow a 16 day repeat cycle designated as the Worldwide Reference System (WRS-2), which provides a global notional gridded mapping for identifying individual Landsat scenes. The Landsat 8 and 9 satellites are flown such that their orbital tracks are separated by 8 days in this 16 day cycle. During the commissioning period of Landsat 9, and during its ascent to its operational WRS-2 orbit, the Landsat 9 satellite's orbital track went under and crossed over the orbital track of the Landsat 8 satellite. This produced a unique situation where nearly time-coincident imagery could be obtained from the instruments of the two spacecrafts. From a radiometric standpoint, this allowed for near-time cross-calibration between the instruments to be performed. From a geometry perspective, calibration is achieved through high-resolution reference imagery over specific ground locations, thus ensuring calibration of the instruments and for the instruments to be well cross-calibrated geometrically. Although these underfly data do not provide calibration of the instruments between the platforms from a geometric perspective, they allow for the verification of the calibration steps involving the instruments and spacecraft. This paper discusses the co-registration of this unique set of data while also discussing other geometric aspects of these data by looking at and comparing the differences in sensor viewing and sun angles associated with the collections from the two platforms for imagery obtained over common geographic locations. The image-to-image comparisons between Landsat 8 and 9 coincident pairs, where both datasets are precision terrain products, are registered to within 2.2 m with respect to their root-mean-squared radial error (RMSEr). The 2.2 m represents less than 0.1 of a 30 m multispectral pixel in misregistration between the L9 and L8 underfly products that will be available to the user community. This unique dataset will provide well-registered, near-coincident image acquisitions between the two platforms that can be a key to any calibration or application comparisons. The paper also presents that, for images for which one of the image pairs failed precision corrections and became a terrain-corrected only product type, a range of 8–14 m RMSEr could be expected in co-registration, while, in cases where both image pairs failed the precision correction step and both images became a terrain-corrected only product type, a 14 m RMSEr could be expected for co-registration. [ABSTRACT FROM AUTHOR]

Textual data such as tags, sentence descriptions are combined with visual cues to reduce the semantic gap for image retrieval applications in today's Multimodal Image Retrieval (MIR) systems. However, all tags are treated as equally important in these systems, which may result in misalignment between visual and textual modalities during MIR training. This will further lead to degenerated retrieval performance at query time. To address this issue, we investigate the problem of tag importance prediction, where the goal is to automatically predict the tag importance and use it in image retrieval. To achieve this, we first propose a method to measure the relative importance of object and scene tags from image sentence descriptions. Using this as the ground truth, we present a tag importance prediction model to jointly exploit visual, semantic and context cues. The Structural Support Vector Machine (SSVM) formulation is adopted to ensure efficient training of the prediction model. Then, the Canonical Correlation Analysis (CCA) is employed to learn the relation between the image visual feature and tag importance to obtain robust retrieval performance. Experimental results on three real-world datasets show a significant performance improvement of the proposed MIR with Tag Importance Prediction (MIR/TIP) system over other MIR systems. [ABSTRACT FROM PUBLISHER]