外文翻译-原文.pdf

扇贝自动分选装置结构设计含CATIA三维图三维及12张CAD图.zip

收藏

资源目录
跳过导航链接。
压缩包内文档预览:
预览图 预览图 预览图 预览图 预览图 预览图
编号:42889724    类型:共享资源    大小:45.25MB    格式:ZIP    上传时间:2020-01-20 上传人:QQ14****9609 IP属地:陕西
400
积分
关 键 词:
扇贝 自动 分选 装置 结构设计 CATIA 三维 12 CAD
资源描述:
扇贝自动分选装置结构设计含CATIA三维图三维及12张CAD图.zip,扇贝,自动,分选,装置,结构设计,CATIA,三维,12,CAD
内容简介:
Proceedings of ISOT 12, Inti. Symposium on Optomechatronic Technologies Bottom Sediment Classification Method from Seabed Image for Automatic Counting System of Scallop Koichiro Enomoto Graduate School of Masashi Toda Center for Multimedia Yasuhiro Kuwahara Hokkaido Research Organization, Fisheries Research Department Abashiri Fisheries Research Institute Abashiri, Japan 099-3119 Email: kuwahara-yasuhirohro.or.jp Systems Information Science, Future University Hakodate Hakodate, Japan 041-8655 Email: g3111001fun.ac.jp and Infonnation Technologies, Kumamoto University Kumamoto, Japan 860-8555 Email: todacc.kumamoto-u.ac.jp Abstract-Each related organization conducts various fishery investigations and collects data required for estimation of resource state. In the scallop culture industry in Tokoro, Japan, the fish resources are investigated by analyzing seabed images. The seabed images are now obtainable from catamaran technology. However, there is no automatic technology to measure data from these images, and so the current investigation technique is the manual measurement by experts. The scallop is looked different from each environment. Therefore, a suitable algorithm to extract the scallop area depends on the bottom sediments of the seabed image. In this paper, we propose a method to classify the bottom sediments of the seabed image. For bottom sediment classification, we forge a strong classifier from weak classifiers using AdaBoost using the various texture features. This paper describes a method to classify the bottom sediments, presents the comparison of the effectiveness of the texture features and the results. Moreover, we presents the experiments results of the scallop counting based on the proposed method, and evaluate the methods effectiveness. I. INTRODUCTION Each related organization conducts various fishery investigations and collects data required for estimation of resource state 1. The results of investigations are used to estimate the catch size, time fish are caught, and future stocks. Therefore, investigations must more accurately measure the numbers, sizes, and states of fisheries. Recently, investigation method have been developed to measure fishes using an underwater camera or its video 2, 3. However, these investigations do not using automatic measurement from underwater camera or its video. T. Hagisawa et al. showed a method to measure the red algae for the fishery investigation of laminaria from underwater video camera 4. In the scallop culture industry in Tokoro, Japan, the fish resources are investigated by analyzing seabed images 1. The seabed images are now obtainable from catamaran technology. However, there is no automatic technology to measure data from these images, and so the current investigation technique is the manual measurement by experts. Therefore, this investigation is neither efficient nor long ranging. Automatic counting systems must be developed to measure the scallop more quickly and investigate fisheries more accurately. The scallop habitat is the seabed of gravel and sand 5, 6, and the distribution of bottom sediments include the these fields in the fishing area, Tokoro 5. K. Enomoto, et al. have already proposed the algorithms to count the scallops from the seabed image of gravel and sand, because the scallop is looked different from each environments 7, 8. They have already proposed the algorithms to extract the scallop areas from the seabed images of gravel and sand, because the scallop is looked different from each environments. Therefore, a suitable algorithm depends on the bottom sediments of the seabed image. In oceanography and fisheries science, the methods of bottom sediment classification have been proposed from the image of side scan sonar (SSS) 9-12. However, there is not a method to classify the bottom sediments from the seabed image. In this paper, we propose a method to classify the bottom sediments from the seabed image. For the bottom sediment classification, we forge a strong classifier from weak classifiers using AdaBoost, and use the various texture features. This paper describes a method to classify the bottom sediments, presents the comparison of the effectiveness of the texture features and the results. Moreover, we present the experiment results using proposed method for the scallop of measurement. The next section describes the seabed images, the scallop, the seabed environments, and design considerations for bottom sediment classification from the seabed image. We describe the various texture features, and propose a method to classify the bottom sediments using AdaBoost in Sec. II. Section III shows the experimental method and the result gained from applying our method to the seabed images, and discusses the validity of our method. A. Seabed Environments and Scallop Figure 1 shows the digital photographs of the seabed fields of gravel and sand (1536 x 1024 pixels and in 24-bit color). 978-1-4673-2877-7112/$31.00 2012 IEEE These seabed images contain scallops, gravel, dead shells, and so on. In Fig. l(a), the gravels are a variety of sizes, shapes, and colors. In Fig. 1 (b), the grain size of sand is smaller than gravel. There are very few the gravel, something of chromatic color in the sand field. Figure 2 shows the scallop areas of 64 x 64 pixels in the fields of gravel and sand. In the gravel field, the scallop areas have special features, such as being sepia or sienna in color and shaped like fans with a striped pattern (Fig.2(a) and (b). The scallop cannot hide under the gravel, because the gravel is big and heavy for its. In the sand field, the scallop areas have special features, such as the shelly rims being white and shaped like fans (Fig.2(c) and (d). There is no sand in the shelly rim, but the scallop shell is covered with sand. The scallop opens and closes its shell while it is alive and breathing. For the same reason, the scallop does not overlap with other scallops in the fields of gravel and sand. These facts are based on the professional knowledge of ecologists and fishennen. A suitable algorithm shold depend on the bottom sediments of the seabed image. K. Enomoto have already proposed the algorithms to extract the scallop areas from the seabed images of gravel and sand, and shown the methods effectiveness 7, 8, because the scallop is looked different from each environments. These extraction rate accuracy are 95.0% 7 and 89.0% 8 in the scallop extraction algorithm for the fields of gravel and sand. Therefore, it is need to select a suitable scallop extraction algorithm on the basis of each field, and develop the bottom sediment classification method from the seabed image. For oceanography, fisheries science, and fishery industry, it is important in the bottom sediment classification. The results of bottom sediment classification is used to estimate the sources of natural resources, tide and fishing place, and important for fisheries management. The fish, seaweed, and fishshell are changed by the bottom sediment in ecology 5. Moreover, the benthic and the bottom sediment influence each other. The methods of bottom sediment classification have been proposed from the image of side scan sonar (SSS) 9-12. These methods classify the bottom sediment using the texture analysis of the image of side scan sonar. L. Atallah and P.J.P. Smith showed a method of the bottom sediment classification using the texture features and K-nearest neighbor 9. H. Lannaya et al. showed also a method using the texture features and support vector machine 12. These methods can not apply the bottom sediment classification of the seabed images, because the seabed image has the different information from the images of side scan sonar. The seabed images are color, and high space resolution more than the images of the side scan sonar. Therefore, the texture features require to be selected for the bottom sediment classification to measure the scallop from the seabed image. (a) (b) Fig. I. Seabed images. (a) Gravel field. (b) Sand field. (a) (b) (c) (d) Fig. 2. The images of scallop area (size:64 x 64). The gravel seabed images are (a) and (b). The sand seabed images are (c) and (d). B. Proposed Method Figure 3 shows our proposed system to extract the scallop area. First, the object image is performed the preprocessing 8. Next, the seabed environment of the object image is identified by the bottom sediment classification. The scallop areas are extracted using a suitable scallop extraction algorithm based on the result of the bottom sediment classification. Finally, the counting results show the user. In this paper, we propose a method to classify the bottom sediments of the seabed image. We forge a strong classifier from weak classifiers using AdaBoost, and use the various texture features such as the histogram, GLCM, and GLDM. Section II explains a method. Section III shows also the results gained from applying our method to the seabed images, and Input image . Preprocessing I Smoothing I I Extract recognizable area I i-t Remove frame area J + Bottom sediments classification I Identify using AdaBoost I Gravel fiel andfield I Extraction algorithm II Extraction algorithm I for the gravel field for the sand field . . Result image Fig. 3. Proposed counting system. J the results of the scallop extraction based on the bottom sediment classification. II. BOTTO M SEDI MENT CLASSIFICATION A. AdaBoost A suitable algorithm to extract the scallop area should depend on the bottom sediments of the seabed image. In this paper, the bottom sediments are separated the fields of gravel and sand. We describe a method to classify the bottom sediments of the seabed image using AdaBoost 13. AdaBoost is a typical instance of Boosting learning. AdaBoost learning algorithm is used to boost the classification performance of a simple learning algorithm. It does this by combining a collection of weak classification functions to form a stronger classifier. In the language of boosting the simple learning algorithm is called a weak leaner. After the first round of learning, the examples are re-weighted in order to emphasize those which were incorrectly classified by the previous weak classifier. The final strong classifier takes the form of a perceptron, a weighted combination of weak classifiers followed by a threshold. The AdaBoost algorithm is shown in Fig.4. In this paper, the examples images(xi Yi) are clipped the seabed images. If the image Xi is the gravel field, Yi = 1, and if the image Xi is the sand field, Yi = -1. In strong classifier H (x), if the value is positive, the input image X is the gravel field, and if the value is negative, it is the sand field. We used the texture features, such as the histogram, the gray level co-occurrence matrix (GLCM), and the gray level difference matrix (GLDM). These features are calculated the each color elements in color space of RGB and HLS. The texture features are described in the following section. B. Weak Learner 1) Color Histogram: There are the various colorful gravels in the gravel field, but there are the gray sands and the white shell debris (Fig. 1). We define the color features as the histogram. In this paper, we used these statistics of the histogram such as, Mean Standard Deviation Contrast. These features are calculated the each color elements in color space of RGB and HLS. The weak leaner h(x) for the statistics of the histogram Feolor are defined as h(x) = +1 -1 if Peolor T, otherwise. (1) Here, T are the thresholds for the each color features. If a statistics Feolor is more than a threshold T, the image is the gravel field, otherwise this is the sand field. 2) GLCM The gray level co-occurrence matrix (GLCM) 14, 15 is a popular statistical method of extracting textural features from images. The values of the co-occurrence matrix elements present relative frequencies with which two neighboring pixels where one of them has gray level i and other j. Such matrix is symmetric and also a function of the angular relationship between two neighboring pixels. The GLCM is computed over a segment, which is inspected by a displacement vector 6, defined by its amplitude r and its orientation e (Fig. 5). In this paper, we used these statistics of the GLCM such as Mean Standard Deviation ASM (Angular Second Moment) Energy Entropy Correlation Homogeneity Dissimilarity Contrast. Here, we set r = I, 3, 5 and e = 0. The weak leaner h(x) for the statistics of the GLCM FCLCM are defined as h(x) = +1 -1 if PCLCM T, otherwise. (2) Here, T are the thresholds for the each features. If a statistics FCLCM is more than a threshold T, the image is the gravel field, otherwise this is the sand field. 3) GLDM: The gray level difference matrix (GLDM) 15, 16 is a statistical method of extracting textural features like GLCM. The values of the difference matrix elements present relative frequencies of difference with which two neighboring pixels where one of them has gray level i and other j. A displacement vector is the same as 6 of the GLCM. Give: (Xl,Yl), . ,(xm,Ym) where Xi E XYi E Y = -1,+1 Initialize Dl (i) = . For i = 1, . , T: Train weak learner using distribution Dt. Get weak hypothesis ht : X -+ -I, + I with error tt = PriD, ht(Xi) i-Yi Choose at = ln( l:tE ). Update: Dt(i) x e-at if ht(Xi) = Yi Zt eat if ht(Xi) i-Yi Dt(i)exp( -atYiht(Xi) Zt where Zt is a normalization factor (chosen so that Dt+1 will be a distribution). Output the final hypothesis: T H(x) = sign(L atht(x). t=l Fig. 4. AdaBoost algorithm. In this paper, we used these statistics of the GLDM such as Mean SD (Standard Deviation) ASM (Angular Second Moment) IDM (Inverse Difference Moment) Entropy Contrast. Here, we set r = I, 3, 5 and e = 0. The weak leaner h(x) for the statistics of the GLDM FCLDM are defined as h(x) = +1 -1 if PCLDM /), otherwise. (3) Here, /), are the thresholds for the each features. If a statistics FCLDM is more than a threshold /) the image is the gravel field, otherwise this is the sand field. III. EXPERI MENTS In this section, we will evaluate the accuracy of the bottom sediment classification, and apply the proposed method to extract the scallop area. A. Method 1) Experiment 1: Recognition Accuracy Evaluation: In the weak leaner, we use the combination of the 7 data sets. The data sets are Data I Histogram Data 2GLCM Data 3GLDM Data 4Histogram and GLCM Data 5Histogram and GLDM (Xy,) Gray value j (x1, Y 1) Gray value j Fig. 5. Definition of relative locations of two pixels. Data 6GLCM and GLDM Data 7 Histogram , GLCM, and GLDM. We examine the accuracy of the bottom sediment classification using the leave-one-out cross validation (LOOCV). Leaveone-out cross validation involves using a single observation from the original sample as the validation data, and the remaining observations as the training data. This is repeated such that each observation in the sample is used once as the validation data. Each data sets are evaluated the accuracy of the bottom sediment classification using LOOCV. We used 240 gravel images and 149 sand images in this experiment. These texture images were selected the upper left of the middle of the frame that have a margin 64 x 64 pixels, and its size is 128 x 128 pixels 8. Figure 6 shows the original samples in the fields of gravel and sand. 2) Experiment 2: Scallop Extraction: We examine the scallop extraction in combination with the proposed method. In the gravel field, the scallops and the other areas are extracted by the scallop extraction algorithm for the sand field. The opposite is also true. We tried to extract the scallop areas using the result of the bottom sediment classification. (a) (b) (c) (d) Fig. 6. The samples of a original samples (size: 128 x 128). The gravel seabed images are (a) and (b). The sand seabed images are (c) and (d). (a) (b) Fig. 7. The samples of a scallop image (size: 128 x 128). The gravel seabed image is (a). The sand seabed image is (b). In the gravel field, the scallop areas are extracted by using following processes 7. First, the candidate scallop areas were extracted only using only shape features (the strong edge points). Some candidate areas were obtained in this process. The scallop area is selected from the candidate areas using the color feature or the striped pattern feature. In the sand field, the scallop areas are extracted by using following processes 8. The candidate shelly rim pixels are extracted from the obtained image by dynamic threshold processing. Next, the candidate scallop areas are extracted using the Hough transform. Finally, the scallop areas are extracted by threshold processing using the shell feature and the shelly rim feature. Here, the image with the scallop area is referred to a scallop image, and the image without this is referred to an other image. We used 24 scallop images and 87 other images in the gravel field, and 20 scallop images and 87 other scallop images in the sand field. Here, we used only the scallop images that all scallops are extracted by the suitable extraction algorithm on the each fields, and the Data 7 for the bottom sediment classification. Figure 7 shows the images of scallop and other in the fields of gravel and sand. B. Results 1) Results of Experiment 1: Table I presents the experiment results of the recognition rate of the bottom sediment classification. All recognition rate is more than 96%. The recognition rates of the gravel and the sand are 100.0% and 96.6% in Data 6 and 7. The samples of the bottom sediment classification are presented in Fig. 8. In Fig. 8(a)-c), these images were recognized incorrectly only a feature such as histogram, GLCM, and GLDM, but its were recognized correctly using these features and AdaBoost. 2) Results of Experiment 2: Table II and III present the experiment results of the scallop extraction using the scallop extraction algorithm for the fields of gravel and sand. All images are recognized correctly in Table II and III. In the TABLE I THE RECOGNITION RATE OF THE BOTTOM SEDIMENT CLASSIFICATION. Recognition rate Gravel Sand Total Data I 98.3% 96.0% 97.43% Data 2 99.6% 97.3% 98.7% Data 3 98.8% 96.0% 97.7% Data 4 99.2% 97.3% 98.5% Data 5 99.2% 96.6% 98.2% Data 6 100.0% 96.6% 98.7% Data 7 100.0% 96.6% 98.7% Fig. 8. The sample images of the bottom sediment classification. The gravel seabed images are (a) and (b). The sand seabed image is (c). These images were not recognized correctly only a texture features, but there were recognized using Data 6 and 7. gravel field, Some images were extracted the areas using the scallop extraction algorithm for the sand field. The examples of the extracted area are presented in Fig. 9. The scallop extraction algorithm for the sand field extracted incorrectly in the gravel field images of scallop and other (Fig. 9). In the scallop images, all extracted areas were error by the scallop extraction algorithm for the sand field. However, in the sand field, there is not the extracted area using the scallop extraction algorithm for the gravel field. TABLE II RESULTS OF EXPERIMENT 2 IN THE GRAVEL FIELD. Scallop Other Num. images 24 87 Scallop extraction Gravel 24 (100.0%) 87 (100.0%) algorithm Sand 6 (25.0%) 39 (44.8%) Bottom sediment Gravel (True) 24 (100.0%) 87 (100.0%) classification Sand (False) 0(0.0%) 0(0.0%) TABLE III RESULTS OF EXPERIMENT 2 IN THE SAND FIELD. Scallop Other Num. images 20 87 Scallop extraction Gravel 0(0.0%) 0(0.0%) algorithm Sand 20 (100.0%) 87 (100.0%) Bottom sediment Gravel (False) 0(0.0%) o (0.0%) classification Sand (True) 20 (100.0%) 87 (100.0%) Fig. 9. Experiment result samples using the scallop extraction algorithm for the sand field in the gravel field. The scallop images are (a) and (b). The other images are (c) and (d). C. Discussion We developed a method of the bottom sediment classification from the seabed images using AdaBoost. In Fig.8(a)(c), these images were recognized incorrectly only a feature such as histogram, GLCM, and GLDM, but it were recognized correctly using these features and AdaBoost. Therefore, the proposed method using AdaBoost is effective, because AdaBoost selected and integrated the classifiers that better identifies. The texture features are best the data set using the histogram, GLCM, and GLDM for the bottom sediment classification. In Experiment 2, there are the extracted areas using the scallop extraction algorithm for the sand field in the gravel field (Fig. 9). the scallop extraction algorithm for the sand field uses the white pixels 8, but the scallop extraction algorithm for the gravel field uses the strong edge points 7. Therefore, the scallop extraction algorithm for the sand field extracted incorrectly the scallop areas in the gravel field (Table II). the scallop extraction algorithm for the gravel field extracted the no areas in the sand field (Table III), because there is very few of the strong edge points and a gravel in the sand field. Therefore, our proposed method is important for the automatic counting system of scallop. It is said that the counting accuracy of experts manual measurement is 95%. These results are accurate enough for the counting of the scallops. There results show also the effectiveness of the proposed method for the scallop counting. IV. CONCLUSION This paper has presented the methods to classify the bottom sediments and extract the scallop areas from the seabed images. This method uses the texture features as histogram, GLCM, and GLDM. We examined the accuracy by the combination. Additionally, the scallop was counting using a suitable algorithm to extract its based on the result of the proposed method. The experimental results gained by applying the proposed method to the seabed images show our method to be useful. Future work, we will develop the application to count the scallop from seabed images, and apply our technology to the videos for the fishery investigation. ACKNOWLEDG MENT This work was supported by JSPS Grant-in-Aid for JSPS Fellows Number 24-11 059. REFERENCES II Hokkaido Abashiri Fisheries Experiment Station, Monitoring Manual. Hokkaido Abashiri Fisheries Experiment Station (2006) 21 N. Honda, T. Watanabe,
温馨提示:
1: 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
2: 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
3.本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
提示  人人文库网所有资源均是用户自行上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作他用。
关于本文
本文标题:扇贝自动分选装置结构设计含CATIA三维图三维及12张CAD图.zip
链接地址:https://www.renrendoc.com/p-42889724.html

官方联系方式

2:不支持迅雷下载,请使用浏览器下载   
3:不支持QQ浏览器下载,请用其他浏览器   
4:下载后的文档和图纸-无水印   
5:文档经过压缩,下载后原文更清晰   
关于我们 - 网站声明 - 网站地图 - 资源地图 - 友情链接 - 网站客服 - 联系我们

网站客服QQ:2881952447     

copyright@ 2020-2025  renrendoc.com 人人文库版权所有   联系电话:400-852-1180

备案号:蜀ICP备2022000484号-2       经营许可证: 川B2-20220663       公网安备川公网安备: 51019002004831号

本站为文档C2C交易模式,即用户上传的文档直接被用户下载,本站只是中间服务平台,本站所有文档下载所得的收益归上传人(含作者)所有。人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私,请立即通知人人文库网,我们立即给予删除!