Research On The Key Technologies Of Spaceborne Remote Sensing Image Real-time Compression

Real-time data transmitting from space to earth is the basic function of most satellites. With the rapid development of optics technology, the resolving power of generic optics remote sensing is 1m and the highest resolving power of it is 0.1m nowadays. If bits of every pixel are 16, the initial data speed of satellite image can reach to several Gbps. However, the capability of transmit channel is only 300 to 600 Mbps. Therefore, it is in urgent need to solve the problem of the real-time compression of initial image data.Supported by the project "Development of xxx satellite data compression equipment" and the 863 National High Technology Research and Development Program "xxxx image data compression technology", this dissertation concentrates its attention on the key technology of Spaceborne remote sensing image data compression. Nowadays remote sensing image coding algorithms are generally based on Discrete Cosine Transform (DCT) or Discrete Wavelet Transform (DWT). Therefore, this dissertation emphasizes on the improvement of Lapped Biorthogonal Transform (LBT) which is based on DCT along with its coding algorithm and the improvement of JPEG2000 based on DWT. Also, linking with special hardware platform based on DSP, the dissertation studies the parallel implementation in DSP of above algorithm and parallel processing.The main contents and contributions of this dissertation are as follows:1. Aiming at the shortcut of precision and computation complexity of binary integer LBT in DSP implementation, a fixed point algorithm of LBT which fits parallel processing in DSP is proposed firstly. Number of Q format is used to approximate the value of trigonometric function of the rotate matrix. The algorithm is improved in the course of truncation. Then, the parallel implementation of fixed point LBT in DSP is studied. By using assignment partition, complex instruction, software pipeline and complier options, the processing speed is improved greatly. Experiment results show that fixed point LBT has better performance and lower computation complexity comparing with binary integer LBT.2. A modified zerotree coding algorithm is proposed. The techniques of zerotree structure's speediness identification and linear index are used in this algorithm. A coder based on Golomb code which is of great advantage to rapid computation in DSP is adopted. The processing speed is improved through task combining, complier options, software pipelining, loop unrolling, and complex instruction. Experiment results show that the algorithm's memory requirement is low and its coding performance is very competitive with SPIHT while its complexity is far lower than SPIHT. 3. A context based arithmetic coding of LBT coefficients is proposed. There are 60 context models in the algorithm. The MQ-coding is isolated from the bit plane scanning and a modified MQ-coder is designed. For the implementation in DSP of the algorithm, the computing task is assigned newly and the data structure is adjusted largely. Experiments show that the algorithm's coding performance is mostly the same as SPIHT and very competitive with JPEG2000.4. A modified JPEG2000 which is of great advantage to parallel processing is proposed. The modification of JPEG includes: fixed point implementation based on post-scaling of two-dimensional 9/7 wavelet, isolation of signal coding in EBC, modification to context model, isolation of MQ-coding from the bit plane scanning and modified MQ-coder. To save memory, the multilevel combinative line-based wavelet transform is proposed. Experiments show that the algorithm's coding performance is very competitive with JPEG2000.5. A satellite image compression hardware platform based on FPGA and parallel multiple DSPs is designed. The hardware architecture has the advantages of strong data processing capability, good parallel performance and easy update characteristic. And it is designed with good dependability. Then the compression system based on LBT + modified zerotree coding algorithm and the hardware platform is tested. It is shown that the system can finish the requiring function at requiring performance standard and condition. The performance of the algorithm is good.