【摘要】： 隨著各種經(jīng)濟、民事案件中利用偽造文件制成時間進行違法犯罪活動的增多,鑒定文件制成時間成為文檢人員急待解決的問題。作為文件真實有效性憑據(jù)之一的印章印文,不再僅僅涉及同一認(rèn)定的問題,越來越多的爭議是關(guān)于蓋印時間,也即蓋印時間是否與文件所標(biāo)稱的時間相一致的問題。根據(jù)印章印文可變性印跡特征鑒別印章的蓋印時間,是解決此類問題的有效途徑之一。 另一方面,由于計算機技術(shù)的飛躍發(fā)展,人工神經(jīng)網(wǎng)絡(luò)在數(shù)據(jù)挖掘領(lǐng)域的廣泛應(yīng)用,本文提出應(yīng)用BP神經(jīng)網(wǎng)絡(luò)技術(shù)對文檢專家識別出的經(jīng)驗數(shù)據(jù)進行挖掘,最終實現(xiàn)印章印文蓋印時間的輔助識別。本文的主要研究內(nèi)容如下: 首先,提出了基于專家經(jīng)驗的印章特征值指標(biāo)體系。通過文檢專家識別的印文特征,分析出影響印章蓋印時間識別的重要的可變性印跡特征,對這些特征進行分析與匯總,得到一套科學(xué)合理的特征值指標(biāo)體系。 其次,在建立基于專家經(jīng)驗的印章特征值指標(biāo)體系的基礎(chǔ)上,對特征值進行量化處理,將定性的印文特征轉(zhuǎn)換為定量的特征,為BP神經(jīng)網(wǎng)絡(luò)的應(yīng)用提供合理可靠的數(shù)據(jù)支持。 最后,論文分析了BP神經(jīng)網(wǎng)絡(luò)進行印章印文蓋印時間識別的原理,利用三層前饋神經(jīng)網(wǎng)絡(luò)建立識別模型,詳細(xì)探討了網(wǎng)絡(luò)的拓?fù)浣Y(jié)構(gòu)、隱含層節(jié)點個數(shù)確定的原則、樣本數(shù)據(jù)的選取和預(yù)處理、初始參數(shù)的確定、激活函數(shù)的選取等問題。用C#語言實現(xiàn)了改進的BP學(xué)習(xí)算法。以某一公司、某一類型的印章為例,建立對應(yīng)的印章蓋印時間識別的神經(jīng)網(wǎng)絡(luò)模型。通過樣本數(shù)據(jù)以及測試數(shù)據(jù)的仿真實驗表明,該模型能夠滿足高精度的要求,具有較好的泛化能力。通過將該模型應(yīng)用到印章蓋印時間識別領(lǐng)域,實現(xiàn)了印章蓋印時間識別的科學(xué)化和自動化,同時表明了應(yīng)用BP神經(jīng)網(wǎng)絡(luò)識別印章蓋印時間的有效性和實用價值。
[Abstract]:With all kinds of economy, the use of forged documents to make time for criminal activities increased in civil cases, and the time of making identification documents has become an urgent problem to be solved by document inspectors. As one of the evidences of the true validity of the document, the seal is no longer involved in the same issue, more and more controversy is about the time of seal, that is, whether the time of seal is consistent with the nominal time of document. One of the effective ways to solve this problem is to identify the time of seal according to the feature of variability imprinting of seal print. On the other hand, due to the rapid development of computer technology and the wide application of artificial neural network in the field of data mining, this paper proposes to use BP neural network technology to mine the empirical data identified by document inspection experts. Finally, the identification of seal time is realized. The main contents of this paper are as follows: firstly, a seal characteristic value index system based on expert experience is proposed. By analyzing the features of print recognized by document inspection experts, this paper analyzes the important variable imprinting features that affect the recognition of seal, analyzes and summarizes these features, and obtains a set of scientific and reasonable characteristic value index system. Secondly, based on the establishment of the seal characteristic value index system based on the expert experience, the characteristic value is quantified, and the qualitative print feature is converted into the quantitative feature, which provides reasonable and reliable data support for the application of BP neural network. Finally, the paper analyzes the principle of BP neural network for seal time recognition, establishes the recognition model by using three-layer feedforward neural network, and discusses in detail the topological structure of the network and the principle of determining the number of hidden layer nodes. The selection and preprocessing of sample data, the determination of initial parameters, the selection of activation function and so on. The improved BP learning algorithm is implemented in C # language. Taking the seal of a certain company and a certain type of seal as an example, a neural network model for the identification of seal time is established. The simulation results of sample data and test data show that the model can meet the requirement of high precision and has better generalization ability. By applying the model to the field of seal time recognition, the scientific and automatic identification of seal time is realized, and the validity and practical value of BP neural network in identifying seal time are also demonstrated.
【學(xué)位授予單位】：大連海事大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2009
【分類號】：D918.91

【相似文獻(xiàn)】

相關(guān)碩士學(xué)位論文 前1條

1 張金源;基于神經(jīng)網(wǎng)絡(luò)的印章蓋印時間識別的研究[D];大連海事大學(xué);2009年

，

本文編號：2344412