The time - frequency of the wavelet transform and the signal wavelet decomposition - reconstruction algorithm based on the multi - resolution analysis are analyzed . the signal local singularity under the wavelet transform are studied . the difference between between mallat ' s maximun modulus algorithm and donoho ' s waveshrink algorithm are analyzed , too . the denoise effect of different threshold and threshold function are studied . base on compare the advantage and disadvantage of donoho & johnstone threshold and believing - region threshold , a modified algorithm is developed : combining threshold distinguishing and shield filter method 討論了基于mallat理論最大模數(shù)方法以及基于donoho理論的閾值萎縮去噪方法的差別,研究比較了不同閾值規(guī)則與閾值函數(shù)對去噪效果的影響;在此基礎(chǔ)上,結(jié)合統(tǒng)一閾值和置信區(qū)間閾值各自的優(yōu)缺點,提出閾值判別和屏蔽濾波相結(jié)合的方法。
Finally we study a modified algorithm , which applies the fast fixed - point algorithm for complex signals to blind signals eparation in frequency domain . it ’ s a one - unit algorithm . to prevent converging to the same signal , the deflation algorithm is used to separate signals one by one 4 .結(jié)合上述卷積混合信號的頻域分離算法,研究了一個改進(jìn)單源算法,將復(fù)數(shù)快速固定點算法應(yīng)用到頻域加速收斂,為了防止算法收斂于同一信號,利用一種抽氣技術(shù),這樣可一次分離出一個信號,實現(xiàn)卷積混合信號分離。
Then , with deeper understanding of them , we hope to make some improvements . in the aspect of object detection and segmentation , we begin with the static background case . two modified algorithms , adaptive time - delayed difference model and mixture background estimation model based on mra , are proposed 面對這種研究現(xiàn)狀,本文試圖在前人工作的基礎(chǔ)上,通過對運動目標(biāo)檢測、分割和識別算法的分類研究,探索各種不同算法內(nèi)在的一致性,并期望在此基礎(chǔ)上增強對算法的理解以及對某些問題的求解提出改進(jìn)。
In chapter 3 , we give an equivalent form of semi - infinite programming , and a locally convergent ssle method is proposed for sip . we only need solve a linear system equations and a subproblem with a parameter per step , also a modified algorithm which saves cost of computations is given , at the end of the paper , we give a proof of the convergence for the algorithms 第三章通過適當(dāng)?shù)淖冃?,得到半無限規(guī)劃問題的一個等價形式,并給出一個局部收斂的序列線性方程組算法,這個算法在每一步,只需求解一個線性方程組和一個帶參數(shù)的非線性子問題,證明了算法的收斂性,同時,給出了一個修正算法,與前面算法相比較,修正算法節(jié)約一定的計算量,同樣具有較好的收斂性。
Thirdly , in the traditional decision tree based ensemble gene mining method , the gene number in the node of the tree is constrained to one . a modified algorithm is proposed , which uses the character of the fldt measure to realize the extension that the number of the gene contained in every node can be any number smaller than n ( n is specified by user ) . the extension cancels the traditional algorithm ’ s constraint and makes the algorithm more flexible and more powerful on the classification 第三,在傳統(tǒng)的基于決策樹的集成基因挖掘算法中,樣本空間中的分類平面必須與坐標(biāo)軸平行,大大限制了該算法的性能,針對此,本論文中提出了一種改進(jìn)算法,該算法利用基因集合的fldt測度的獨有特性,實現(xiàn)了節(jié)點包含的基因數(shù)目可以是小于n ( n的值由用戶指定)的任意數(shù)目的改進(jìn),提高了原算法的靈活性和分類能力。
In chapter 3 , a congestion control algorithm facc based on active networks and its modified algorithm mfacc are proposed to solve the problem that congestion control in traditional network depends on the ambiguity implication to detect network congestion and then carry out the congestion control 第三章針對tcp擁塞協(xié)議中存在的受傳統(tǒng)網(wǎng)絡(luò)服務(wù)能力限制,擁塞控制機制只能進(jìn)行隱式的擁塞檢測這一問題,本文提出了基于主動式網(wǎng)絡(luò)的主動擁塞控制算法facc以及它的改進(jìn)算法nfacc 。
The original and modified algorithms are compared on many standard benchmarking datasets , including face detection , usps , and mnist handwritten digit recognition problems . numerical results show that the modified algorithm out - performs the original algorithm with respect to training tune in most cases , especially when the training data are noisy and overlapping 人臉檢測、 usps ,和mnist手寫數(shù)字識別等數(shù)據(jù)的實驗結(jié)果顯示改進(jìn)的算法在很多情況下優(yōu)于原來的算法,尤其是在強噪聲和類間重疊的數(shù)據(jù)下這種改善更為明顯。
It has been widely applied to image and speech compression . the paper deals with image compression based on vector quantization , detailedly expounds its elemental principle , relative conception , and present development , deeply explores its two key technique - - codebook generation and codeword searching , summarizes and analyzes present typical algorithms , and presents modified algorithms 本文以矢量量化技術(shù)在圖像壓縮方面的應(yīng)用作為研究目標(biāo),詳細(xì)闡述了矢量量化的基本原理、相關(guān)概念及發(fā)展現(xiàn)狀,深入探討了矢量量化的兩大關(guān)鍵技術(shù)? ?碼書生成和碼字搜索,總結(jié)分析了現(xiàn)有典型的算法,并提出改進(jìn)算法。
In order to improve the generalization ability and to make the system more accordant to the practical requirement , the paper introduces the k - l information distance to the error function of the traditional bp algorithm . the simulation proves that the generalization ability of the system trained with the modified algorithm is much better than that of other algorithms 為了進(jìn)一步改進(jìn)系統(tǒng)的泛化能力,使其滿足報價的實際需要,本文利用神經(jīng)網(wǎng)絡(luò)的概率描述,通過研究k - l信息距離和神經(jīng)網(wǎng)絡(luò)泛化能力的關(guān)系,構(gòu)造了一個新的神經(jīng)網(wǎng)絡(luò)學(xué)習(xí)誤差函數(shù),并將此法與其它各種算法的泛化結(jié)果進(jìn)行比較。
But it is not suitable for online implementation or for tracking signal components with characteristic that change with time , a new modified algorithm is present based on the adaptive short - time kernel . simulations show that adaptive short - time kernel time - frequency distribution has better performance than anyother time - frequency distribution not only in the time - frequency resolution but also in the cross - components suppression 通過對仿真信號計算其自適應(yīng)短時高斯核時頻分布,仿真結(jié)果表明,無論從改善信號的時頻分辨率還是抑制交叉項方面,自適應(yīng)短時核時頻分布均表現(xiàn)出了較好的特性。