Invoice reimbursement processing method and device

[0029] In order to make the objects, technical solutions, and advantages of the embodiments of the present invention, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, a schematic embodiment of the present invention and a description thereof are for explaining the invention, but is not limited to the present invention. It should be noted that the features in the embodiments and embodiments in the present application may be any combination with each other in the case of an unable conflict.

[0030] figure 1 Is a flow diagram of the processing method provided reimbursement invoices first embodiment of the present invention, as figure 1 Shown, invoices reimbursement processing method according to an embodiment of the present invention, comprising:

[0031] S101, the receiving reimbursement request sent by a client, the request including invoices, pictures and reimbursement employee identification;

[0032] Specifically, when the employees need to obtain reimbursement, reimbursement can be sent by the client request, the server receives the request for reimbursement, the reimbursement request including invoices, pictures and employee identification. Wherein, the electronic invoice may be invoice picture images may be images of paper invoices, not limited in embodiments of the present invention. Employees have a unique corresponding employee identification. The client including but not limited to desktop computers, laptop computers, tablet computers, smart phones and the like. Execution subject invoices reimbursement processing method according to an embodiment of the present invention include, but are not limited to the server.

[0033] For example, a picture can be obtained by scanning paper invoices way.

[0034] S102, the invoice picture regionalization obtain respective regions to be recognized;

[0035] In particular, the server after receiving the invoice image, the invoice may be regional image processing to obtain the various areas to be identified, each zone has to be identified needs to be information for identifying.

[0036] For example, the position of the respective regions can be set in advance to identify certain areas of the invoice, based on the area position of the respective regions to be recognized, to obtain the respective regions to be recognized. Region above position, by an endpoint coordinate of the rectangular area, the length and width of the rectangular region is represented as a rectangular region. Expressed as the above-described end point coordinates (x, y), x may be preset image representing an invoice and a length L y is the percentage of the occupied percentage of the width W of the invoice image b, x = La, y = Wb.

[0037] E.g, figure 2 It is a schematic diagram of an invoice to be identified region provided by the second embodiment of the embodiment of the present invention, as figure 2 , The invoice to be recognized as a thick line area indicated in black rectangular box, including invoice code area is located, the region where the invoice number, the region where the billing date, the name of the region where the purchaser, the purchaser taxpayer identification area where the number of regional goods or taxable services, the service name is located, the total ad valorem tax (capital) region corresponding to the total ad valorem tax area, an area corresponding to the name of the seller's (lowercase), the seller taxpayer identification number area is located, the area where the invoice stamp.

[0038] S103, and the recognition area according to the respective invoices to be recognition model, the recognition result is obtained for each region to be recognized; wherein said invoice an invoice recognition model is based on the corresponding training samples and the training mark information obtained;

[0039] In particular, the server respective regions to be recognized are input to the invoice identification model, identification through invoice identifying the model, the recognition result can be obtained for each area to be identified. Wherein said invoice an invoice recognition model is based on training samples and the flag information corresponding to the obtained training.

[0040] For example, for the area invoice number is located, the recognition result for a specific invoice number; for zone purchaser name is located, the recognition result of the specific name of the purchaser; to the sum (upper) region corresponding to the recognition result monovalent tax amount Chinese capital; for the area where the invoice stamp, invoice stamp picture recognition results.

[0041] S104, if it is determined to be informed of the recognition result of each region identified through the identification check, employee registration information is obtained according to the employee identifier;

[0042] In particular, after obtaining said server identification result for each area to be recognized, the recognition result will be identified for each region identification check result, if the recognition result of each region have to be identified by checking the recognition result, the server will then obtain information according to the registration staff employee ID, employee registration information is the subsequent claims fill invoice information units required. Wherein, the registration information may include information staff employee department, employee name, coded by the employee, the employee payroll accounts, wages and the like Bank account, according to actual needs, the invention is not limited in the present embodiment.

[0043] S105, based on the invoice information, registration information, and the employee claims singlemode Edition, generate invoices expense report; wherein the invoice information is obtained from the recognition results of the respective regions to be recognized.

[0044] Specifically, information, since not all of the recognition result of the recognition area to be filled are required to obtain reimbursement list, the server may need to be obtained to fill a single invoice claims recognition result from the respective regions to be recognized as the invoice information, to fill reimbursement single-mode version, and the employee registration information to fill in single-mode version reimbursement to generate a single invoice claims, realized autofill single invoice claims. Wherein the single-mode version of the reimbursement is the default.

[0045] For example, Table 1 is a reimbursement single mode, the server will acquire the current time as the date of the invoice reimburse, fill in the department of the employee to the application department column, fill in the name bar of the employee, and fill in the employee When you go to the staff coding column, the identification result of the area where the item or taxable service, the service name is located, and the identification result of the area corresponding to the value tax (capital) is filled in the amount (capitalization) column, and the price tax (Capital) The identification result of the corresponding area is filled in the amount (lowercase) column, filling the employee name to the payeer column, filling the employee's salary account to the account bar, filled out the account opening bank of the employee's payroll. In the identification result of the area, the identification result of the area where the service name is located, obtain the keyword to the summary bar, and fill in the number of invoice numbers in accordance with the number of invoices included in the reimbursement request, the identification result of the area where the invoice number is located Fill in the invoice number column to complete the filling of the invoice reimbursement form.

[0046] Table 1 Reimbursement Single Mode Delivery Unit

[0047]

[0048] The processing method of the invoice reimburse provided by the embodiment of the present invention can receive a retribution request transmitted by the client, and the invoice picture is registered, obtaining each pending area, and obtains each pending area and invoice recognition model to obtain each pending The identification result of the area, after judging the identification result of each to be identified by the identification result, obtaining the employee registration information according to the employee identification, according to the invoice information, employee registration information, and the reimbursement single mode version, generate the invoice reimbursement order, and realize the invoice Automatic completion of the reimbursement, improve the generation efficiency of the invoice reimbursement form.

[0049] image 3 It is a flow diagram of the processing method of the invoice reimburse provided by the third embodiment of the present invention, such as image 3 As shown, on the basis of the above-described embodiments, further, the process of invoicing training samples and corresponding tag information training includes:

[0050] S301, get invoice training samples, the invoice training sample includes a preset number of sheets of ticket training;

[0051] Specifically, the preset number of the same type of invoice picture is collected as an invoice training sample. The server can obtain the invoice training sample, and each invoice picture in the invoice sample is called an invoice training picture. Among them, the same type of invoice picture can be an ordinary invoice or a VAT special invoice, according to the actual needs, the embodiment of the present invention is not limited.

[0052] S302, for each invoice training image in the invoice training sample, to obtain a region of each invoice training image, and a mark information corresponding to each of the area to be identified;

[0053] Specifically, the server performs regional processing of each invoice training image in the invoice training sample, and can obtain a region of each invoice training image. For each invoice training picture, each invoice training image can be manually identified, and each invoice training picture is obtained, each invoice training image is marked, the server can be Get each invoice training image of the mark information to be recognized. Among them, the specific acquisition process of each invoice training picture is similar to the specific obtaining process of each to-recognition area of ​​the invoice picture in step S102, which is not described herein.

[0054] For example, for the area where the invoice code is located, the corresponding tag information is a specific string of the invoice code; the corresponding tag information is the specific number of the invoice number; for the invoice date, the corresponding tag information For the specific time of the invoice date; for the area where the purchaser name is located, the corresponding tag information is a corporate name or reimburse name; for the area where the purchase party taxpayer identification number is located, the corresponding mark information is the specific taxpayer identification number; For the area where the goods or taxable services, the service name is located, the corresponding marking information is a statement including keywords such as heating, heating, heating fees, natural gas, electricity bill; for the region, corresponding to the price tax, corresponding to the region, corresponding The marker information is the RMB uppercase of the invoice; for the area corresponding to the price tax (lowercase), the corresponding mark information is the renminbi lowercase of the invoice; the area where the salesperson is located, the corresponding mark information is the company name; for the sales party The area where the taxpayer identification number is located, the corresponding mark information is a specific taxpayer identification number; for the area where the invoice special chapter is in the area, the corresponding mark information is the company's invoice special chapter picture.

[0055] S303, according to the original training model, each invoice training picture to be identified by the area and each of the tag information corresponding to each of the to be identified, the training is obtained, and the original training model uses an improved SSD algorithm model, the improvement The convolution layer of the SSD algorithm model includes three convolutionary cores arranged in sequentially arranged, and the magnitude of the conjunction in sequence is 1 × 1, 3 × 3 and 1 × 1, and the three sequentially arranged volumes. Replace the original 3 × 3 convolved nuclear.

[0056] Specifically, the original training model can be trained to the original training model based on the labeling information corresponding to each invoice training picture and the mark information corresponding to each of the to be identified, and the invoice recognition model can be trained. Wherein, the original training model uses an improved SSD (SINGLE SHOT Detection) algorithm model, the convolution layer of the improved SSD algorithm model comprises three convolutionary cores arranged in sequential, three sequentially arranged volumes. Sex 1 × 1, 3 × 3 and 1 × 1. The improved SSD algorithm model is obtained by replacing an existing SSD algorithm model with a convolution layer of the convolution layer of the existing SSD algorithm model to replace the convolution layer of the existing SSD algorithm model.

[0057] The SSD algorithm model belongs to a convolutional neural network, typically employ VCG-16 network architecture, including 13 convolution layers and 3 full connecting layers. The convolution layer extracts the feature of the input data through the convolutionary core, and each element in the convolutionary core corresponds to a weight coefficient and a deviation value. Each neuron in the convolution layer is connected to a plurality of neurons of the region similar to its position in the front layer, depending on the size of the convolutionary nucleus, and this area is referred to as a perceptual field.

[0058] In convolutional neural networks, the weight parameters required for convolution calculations are concentrated in the convolution layer, resulting in a very time consuming. In the convolution network, the number of convolution parameters in each layer of convolution layer is:

[0059] P = C × kh × kw × c

[0060] Among them, C is the number of channels of the civi core in the convolution layer, and KH is high, KW is the width of the convolutionary nucleus. The convolution layer of the conventional SSD algorithm model is used to convolution calculations, and the number of parameters required is:

[0061] P = C × 3 × 3 × c = 9c 2

[0062] In the embodiment of the present invention, the improved SSD algorithm model replaces the replacement of a 3 × 3 convolution of the original convolution layer into three convolutionary cores, three sequentially arranged coil cores It is 1 × 1, 3 × 3 and 1 × 1, and the method of stacking small size convolutionary core is replaced with large size convolution. In actual use, set the number of channels of the first 1 × 1 coil core of 1 × 1 C, set the number of channels of 3 × 3 convolved cores to C / 2, set the second 1 × 1 convolutionary core. The number of channels is C / 2, then the three convolutionary cores of 3 × 3 convolution are replaced, and the number of parameters required is:

[0063]

[0064] Suppose the size of the image is 28 × 28, which can be consolidated using a volume of 3 × 3, and the length is 1, and the filler is 0, the size of the perception is:

[0065]

[0066] For the same 28 × 28 input pictures, three convolution kernels replaced with 3 × 3 convolution are used, for the first 1 × 1 convolution, the size of the field is:

[0067]

[0068] For the next 3 × 3 coil core, the size of the field is:

[0069]

[0070] For the second 1 × 1 convolulence, the size of the perception is:

[0071]

[0072] The improved SSD algorithm models provided by the embodiments of the present invention coincide with the perceptual field size of the conjunction with the replacement of 3 × 3 convolutionary core. 3 × 3 convolutionary nuclear decomposes a three-layer convolutionary nucleus, thereby obtaining more scales, combining these features, and the effect is better than the characteristics of a single volume annotation.

[0073] The above-described improved SSD algorithm model three sequentially arranged convolutions have added a nonlinear expression ability of the network, increasing the nonlinear expression ability of the network, improves the feature extraction Effect and play an implicit regularization. Large-size convolutional copies make the amount of calculation, not suitable for models with high model depth, and calculating performance will decrease. Using multiple small volume nuclear stacks can reduce the number of parameters, reduce the amount of calculation, increase the speed of operation, and increase nonlinear effects.

[0074] Table 2 Model training comparison results

[0075]

[0076] In order to verify the effect of the improved SSD algorithm model, on the same experimental platform (SUSE12 computer, GPU is NVIDIA TESLA P40), the PASCAL VOC training set is used to train the YOLO algorithm model, SSD algorithm model and improved SSD algorithm model, respectively. The respective corresponding models obtained are detected, and the detection accuracy and speed of the record are shown in Table 2. The YOLO algorithm model is the fastest speed, which can reach 16.1 frame per second, but its detection accuracy is only 58.2map, which cannot meet high precision requirements. The detection accuracy of the SSD algorithm model can reach 79.1map, but its detection speed is slower, only 5.5fps, low efficiency, and does not meet the requirements of real-time detection. The improved SSD algorithm model proposed by the present invention has a detection accuracy of 78.8 mAP, and the detection speed is 10.2 frames per second, and the target can still be accurately detected while improving the detection speed.

[0077] Based on the above embodiments, further, the fourth convolution layer of the improved SSD algorithm model includes the three convolutionary convolutionary cores.

[0078] Specifically, an existing SSD algorithm model can be replaced with a 3 × 3 convolutionary core of the fourth convolution layer of the existing SSD algorithm to replace the three convolved convolution, three convolutionary convolutions The size is 1 × 1, 3 × 3 and 1 × 1 in turn. Among them, the number of channels of the original 3 × 3 coil core is c, which can be set to the number of channels of the first 1 × 1 convolutionary core of the first 1 × 1, the number of channels of the 3 × 3 convolutionary core. C / 2, the number of channels of the second 1 × 1 convolution is C / 2.

[0079] On the basis of the above embodiments, further, in the training of the model, the confidence is greater than the a prior case of the confidence threshold generates a prediction box; wherein each probatation frame corresponds to a confidence.

[0080]In particular, the improved SSD algorithm for training the model, each block corresponds to a priori confidence level, confidence is the confidence of the prediction frame corresponding to. Block may be based on a priori confidence threshold screening, a priori block is not greater than the confidence threshold is not decoded, i.e., will not be used to generate a prediction block, to reduce the amount of calculation and the subsequent non-maximum suppression algorithm linear ( Non-Maximum Suppression, referred to as NMS) iterations, improve training efficiency model. Wherein the confidence value is a preset threshold.

[0081] For example, in the SSD algorithm, a size of m × n wherein FIG mn total cells, the number of each cell block prior to k. If the Y-detection target categories, then generating the prediction block, a total of (Y + 4) kmn prediction value required. During training, a prediction value obtained by the prior frame and the bounding box, the bounding box is the converted value with respect to the prior frame. Wherein prior frame is precalculated, different scales and fixed size frame, which borders the real distribution is very close. In VGG16 network structure, wherein the output of the prior frame in FIG conv4_3, fc7, conv6_2, conv7_2, conv8_2, conv9_2 generation, which produced six FIG feature sizes are 38 × 38,19 × 19,10 × 10, 5 × 5,3 × 3,1 × 1, the number of six in each prior block center point of each generated respectively 4,6,6,6,4,4, so each layer takes on a six wherein FIG priori produced a total of 8732 frames. Frame position information is provided prior to d = (d cx , D cy , D w , D h ), Which corresponds to the bounding box b = (b cx , B cy , B w , B h ), L = (l cx L cy L w L h ), The position of the frame a priori prediction value corresponding to the bounding box. Bounding box needs to be encoded prior frame (the Encode) process, the subsequent predicted, a priori need to reverse the encoding process block bounding box, decoding (the Decode), worth from the prediction information of the target to the bounding box.

[0082] Screening frame by setting a priori confidence threshold, the frame is not greater than the a priori confidence threshold is not decoded, the process of reducing the above-described encoding and decoding. In the process of generating the predicted block, it is necessary to use NMS, since the NMS will block a priori confidence of different iterative comparison, reduces the number of frames used to generate a prediction block by setting a priori confidence threshold, thereby reducing the number of iterations improve the efficiency of obtaining a predicted frame.

[0083] Figure 4 Is a flow diagram of the processing method provided reimbursement invoices fourth embodiment of the present invention, as Figure 4 Shown, based on the above embodiments, further, if the individual is determined to be informed of the recognition result by the recognition area recognition result verification comprising:

[0084] S401, the recognition result is determined to be each identified region satisfies a region corresponding to the invoice validation rules; wherein said invoice validation rules are preset region;

[0085] In particular, the server determines the recognition result to be identified for each region satisfies region corresponding invoice validation rules, and records the determination result for each area to be identified, the determination result is satisfied or not satisfied, showed to be identified region satisfies the recognition result satisfies region corresponding invoice validation rules, the recognition result does not satisfy the show to be recognized does not satisfy the region corresponding to the region of invoice validation rules. Wherein the region invoice validation rules are preset embodiment is not limited to actual needs, according to the embodiment of the present invention.

[0086] For example, for the area where the invoice code recognition result, the number of bits corresponding to the invoice code area invoice validation rules may include a recognition result is equal to a first predetermined value, if the invoice code bits equal to the first preset value , then the determination result of the area where the invoice code to satisfy, if the invoice code bits not equal the first preset value, then the determination result of the area where the invoice code is not satisfied. Wherein the first predetermined value is set according to actual needs, the invention is not limited in the present embodiment.

[0087] For regions where the invoice number recognition result, the number of bits corresponding to the invoice number invoice validation rules may include a region of the recognition result is equal to a second predetermined value, if the number of bits equal to the second predetermined invoice value, the determination result of the region located to meet the invoice number, invoice number and if the number is not equal to a second predetermined value, then the determination result of the area where the invoice number is not satisfied. Wherein the second predetermined value is set according to actual needs, the invention is not limited in the present embodiment.

[0088] For areas where the billing date of the recognition result, the area corresponding invoice validation rules can identify whether or not to include the results of this year or last year, including this year or if the recognition result of last year, the result of the determination of the area where the billing date to meet, if recognition the results do not include this year and last year, the result of the determination of the area where the billing date is not satisfied.

[0089] For the name of the region where the purchaser of the recognition result, the area corresponding invoice validation rules can be reimbursed whether the name of the enterprise or reimbursement to recognize the results of the name, if it is judged to be reimbursed business name or business name reimbursement, then buy judgment results area of ​​the name of the party is satisfied, if the result is not the name of the enterprise or reimbursement reimbursement name, then the result of the determination area of ​​the name of the buyer is not satisfied.

[0090] For areas where the purchaser taxpayer identification number of the recognition result, the area corresponding invoice validation rules can be reimbursed if the taxpayer identification number of enterprises as the recognition result, taxpayer identification number if the result is to sell the business, then the purchaser tax judgment results area where the identification number of people to meet, if it is judged not sell corporate taxpayer identification number, then the result of the determination of the area where the buyer taxpayer identification number is not satisfied.

[0091] For taxable services, the service name of the cargo area of ​​the recognition result, the area corresponding invoice validation rules can be preset to identify whether the results include keywords, if the recognition result includes a preset keyword, goods or taxable services, the judgment result of the service area name where to meet, if the result does not include a preset keyword, judgment results area of ​​goods or taxable services, the name of the service is not satisfied. The preset keywords may heating costs, heating costs, heating costs, gas, electricity, etc., according to actual needs, the invention is not limited in the present embodiment.

[0092] For recognition results total ad valorem tax (capital) corresponds to the region, the region corresponding invoice validation rules may include whether non-Chinese uppercase character recognition result, if the recognition result including non-Chinese uppercase characters, then the total ad valorem tax (capital) corresponding region determination result is not satisfied, if the recognition result does not include non-Chinese uppercase characters, then the total price of the tax determination result (upper) region corresponding to satisfy.

[0093] Recognition result monovalent total tax (lower case) corresponding to a region, the region corresponding to the invoice validation rules may include non-recognition result whether the digits, if the recognition result includes Arabic numeral, the total price of the tax (lower case) corresponding to the region the determination result is not satisfied, if the recognition result does not include the Arabic numeral, the total price of the tax determination result (lower case) corresponding to a region to meet.

[0094] For the name of the region where the seller of the recognition result, the area corresponding invoice validation rules can include the business name recognition result matches the blacklist of business name, business name and the blacklist if the result is included in the business name match, then the result of the determination of the area where the seller is not satisfied with the name, business name and the blacklist if the result is included in the company name does not match, then the judgment result of the area where the name of the seller to meet.

[0095] For sellers region where the taxpayer identification number recognition result, the number of bits corresponding to the taxpayer identification number of the invoice validation rules may include a region of the recognition result is equal to a third predetermined value, and if the number of taxpayer identification number equal to the third predetermined value, then the determination result of the area of ​​the seller where the taxpayer identification number in order to meet, if the number of bits taxpayer identification number is not equal to a third predetermined value, then the region where the seller's taxpayer identification number judgment result is not satisfied. Wherein the third predetermined value is set according to actual needs, the invention is not limited in the present embodiment.

[0096] For areas where the invoice stamp of recognition result, the area corresponding invoice validation rules can be included in the recognition result invoice stamp picture is complete, whether to include "special invoice seal" words, if the recognition result includes a complete picture of invoice stamp and include "special invoice seal" the words, then the result of the determination of the area where the invoice stamp is met, if the recognition result included invoice stamp picture is incomplete or does not include "special invoice seal" the words, then the area where the invoice stamp the result of the judgment is not satisfied.

[0097] S402, if the recognition result is known to be recognized in all regions corresponding to satisfy invoice validation rules area, it is determined that the recognition result of each region to be identified through the identification check.

[0098] In particular, the server after obtaining the determination result of the respective regions to be recognized, if the determination result of each identified region is to be satisfied, the identification results of all instructions to be identified region satisfies region corresponding invoice validation rules, then it is determined each recognition result to be the recognition result by the recognition area check.

[0099] On the basis of the above embodiments, further, the process comprising the regionate invoice image:

[0100] Pretreatment of the invoice picture to enhance the sharpness of the invoice picture.

[0101] Specifically, in order to improve image sharpness invoice, the server may be pre-invoice the picture. Wherein said preprocessing algorithm employed in accordance with actual needs, the invention is not limited in the present embodiment.

[0102] For example, the correction may be made to reinforcing invoice image; wrinkles can be removed invoice picture; by binarization process, highlighting the profile in the invoice text image.

[0103] On the basis of the above embodiments, further, the invoice processing method according to an embodiment of the present invention further comprises reimbursement:

[0104] Returning the invoice retribution form to the client.

[0105] Specifically, the server generates an invoice after expense report, the invoice may be sent to the claims single client, employees can obtain reimbursement collating the single client. After checked, the invoice can be reimbursed for review to provide a single financial officers, financial officers, after approval, you can print invoices for reimbursement list, find the relevant leaders to sign.

[0106] Figure 5 It is a schematic diagram invoices for reimbursement processing apparatus according to an fifth embodiment of the present invention, as Figure 5 , The processing apparatus according to claims invoices include a receiving module 501, regionalization module 502, identification module 503, verification module 504 and the module 505 generates embodiment of the present invention, wherein:

[0107]The receiving module 501 is configured to receive a retribution request transmitted by the client, the reimbursement request including an invoice picture and an employee identifier; the regionalization module 502 is used to process the invoice picture to obtain each of the penetrasts; the identification module 503 is used The identification result of each to be identified is obtained according to each pending area and the invoice recognition model; wherein the invoice recognition model is based on invoice training samples and corresponding marking information training; check module 504 is used for judgment The identification result of each pending area is obtained by the identification result check, according to the employee registration information according to the employee identifier; generating module 505 is used to generate an invoice reimburse according to the invoice information, the employee registration information, and the reimbursement single mode. The invoice information is obtained from the identification result of each pending area.

[0108] Specifically, enterprise employees can send a retribution request through the client, and the receiving module 501 receives the packet request, the reimbursement request including an invoice picture and an employee identifier. Wherein, the invoice picture can be a picture of an electronic invoice, or a picture of a paper invoice, and the embodiment of the present invention is not limited. Enterprise employees have the only corresponding employee logo. The client includes, but is not limited to, desktops, laptops, tablets, smartphones, etc.

[0109] After receiving the invoice picture, the regionalization module 502 can process the invoice picture to obtain each pending area, each of which is required to be identified.

[0110] The identification module 503 inputs the respective pending regions into the invoice recognition model, and the identification of the invoice recognition model can obtain the identification result of each to be recognized. The invoice recognition model is obtained based on invoice training samples and corresponding tag information training.

[0111] After obtaining the identification result of each pending area, the check module 504 recognizes the result check of the identification result of each peer to be recognized. If the identification result of each peer to be identified passes the identification result verification, then the school The test module 504 will obtain employee registration information based on the employee identifier, the employee registration information is the information required for subsequent filling the invoice reimbursement order. Among them, the employee registration information may include information such as employee department, employee name, employee code, employee's salary account, account opening of the employee's salary account, according to actual needs, the embodiment of the present invention is not limited.

[0112] Since the identification result of all the area to be identified is needed to be filled in the invoice reimbursement, the generation module 505 can obtain the information required by the invoice reimbux as the invoice information from the identification result of each pending area, fill in the reimburse mode. And fill in the employee registration information into the reimbursement single mode, thereby generating an invoicing reimbursement, and implements the automatic filling of the invoice reimbursement form. Among them, the reporting single mode is preset.

[0113] The processing apparatus provided by the invoice reimburse according to the embodiment of the present invention can receive a retribution request transmitted by the client, and the invoice picture is registered, obtaining each pending area, and obtains each pending area and invoice recognition model to obtain each pending The identification result of the area, after judging the identification result of each to be identified by the identification result, obtaining the employee registration information according to the employee identification, according to the invoice information, employee registration information, and the reimbursement single mode version, generate the invoice reimbursement order, and realize the invoice Automatic completion of the reimbursement, improve the generation efficiency of the invoice reimbursement form.

[0114] Image 6 It is a schematic structural diagram of the processing apparatus of the invoice reimburse provided by the sixth embodiment of the present invention, such as Image 6 As shown, on the basis of the above embodiment, the processing apparatus of the invoice reimburse according to the embodiment of the present invention further includes acquisition module 506, processing module 507, and training module 508, wherein:

[0115] The acquisition module 506 is used to obtain invoice training samples, which includes preset number of sheets of ticket training pictures; processing module 507 is used to regroup each invoice training image in the invoice training, obtain each piece The invoice training picture is to be identified and the corresponding tag information corresponding to each of the regions to be identified; the training module 508 is used to obtain the area of ​​the region and each invoice training picture to be identified and each of the tag information corresponding to the area to be identified. An invoice recognition model; wherein the original training model employs an improved SSD algorithm model, the convolution layer of the improved SSD algorithm model includes three convolutionary cores arranged in sequential, three sequentially arranged coil cores It is 1 × 1, 3 × 3 and 1 × 1, the three arranged convolution nucleus replace the original 3 × 3 convolutionary core.

[0116] Based on the above embodiments, further, the fourth convolution layer of the improved SSD algorithm model includes the three convolutionary convolutionary cores.

[0117] On the basis of the above embodiments, further, in the training of the model, the confidence is greater than the a prior case of the confidence threshold generates a prediction box; wherein each probatation frame corresponds to a confidence.

[0118] Figure 7 It is a structural diagram of the processing apparatus of the invoice reimburse provided by the seventh embodiment of the present invention, such as Figure 7 As shown, the calibration module 504 includes a determination unit 5041 and a determination unit 5042, wherein:

[0119] The determination unit 5041 determines whether or not the identification result of each to be identified satisfies the corresponding invoice area check rule; wherein the invoice area check rule is preset; determining unit 5042 is satisfied if the identification result of the area to be identified is satisfied. The corresponding invoice area check rule, determine the identification result of each pending area through the identification result check.

[0120] Based on the above embodiments, further, the regionalization module 502 is specifically used for:

[0121] Pretreatment of the invoice picture to enhance the sharpness of the invoice picture.

[0122] Based on the above embodiments, further, the processing apparatus of the invoice reimburse according to the embodiment of the present invention further includes a transmission module, wherein:

[0123] The transmitting module is configured to return the invoice rehead or to the client.

[0124] An embodiment of the apparatus provided by the embodiment of the present invention can be used to perform the processing flow of each method embodiment, and the function is not described herein again, and detailed description of the above method embodiments may be described.

[0125] It should be noted that the processing method and apparatus of the invoice reimburse according to the embodiment of the present invention can be used in the financial sector, or may be used in any of the technical fields other than the financial field, and the application method and apparatus of the embodiment of the present invention The field is not limited.

[0126] Figure 8 It is a schematic structural diagram of an electronic device according to an eighth embodiment of the present invention, such as Figure 8 As shown, the electronic device can include a processor 801, a communication interface 802, a memory 803, and a communication bus 804, wherein the processor 801, a communication interface 802, a memory 803 via a communication bus 804. Complete communication between each other. Processor 801 can call logic instructions in memory 803 to perform the following method: receive a retribution request transmitted by the client, the reimbursement request includes an invoice picture and an employee identifier; a regionalization of the invoice picture, obtaining each pending Region; according to each pending area and invoice recognition model, the identification result of each to be identified is obtained; wherein the invoice recognition model is based on invoice training samples and corresponding tag information training; if it is judged to know the respective pending area The identification result is verified by the identification result check, according to the employee's identification, the employee registration information is obtained; according to the invoice information, the employee registration information, the reimburse mode, generates an invoice reimbursement order; wherein the invoice information from each pending area The identification result is obtained.

[0127] Further, the logic instructions in the above-described memory 803 can be implemented in a computer-readable storage medium when the software function unit is implemented and used as a separate product sales or in use. Based on this understanding, the technical solution of the present invention essentially ors a portion of the prior art or a portion of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including Several instructions are used to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or some steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, RAD-ONLY MEMORY), random access memory (RAM, RANDOM Access Memory), disk or disc or optical discs can store program code .

[0128] This embodiment discloses a computer program product that includes a computer program stored on a non-transitory computer readable storage medium, the computer program including program instructions, when the program instruction is executed by the computer, the computer The method provided by each method embodiment can be performed, including: receiving a retribution request transmitted by the client, the reimbursement request comprising an invoice picture and an employee identifier; a regionalization process for the invoice picture, obtaining each pending area; According to the respective pending area and invoice recognition model, the identification result of each to be identified is obtained; wherein the invoice recognition model is based on invoice training samples and corresponding tag information training; if it is judged to know identification of each to-recognition area As a result, by the identification result check, the employee registration information is obtained according to the employee identifier; according to the invoice information, the employee registration information, the reimburse mode, the invoice reimbursement order is generated; wherein the invoice information from the identification of each pending area. Resulting.

[0129] This embodiment provides a computer readable storage medium, the computer readable storage medium storage computer program, the computer program, which enables the computer to perform the method provided by each method embodiment, including: receiving client transmission. Reimbursement request, the reimbursement request includes an invoice picture and an employee logo; a regionalization process for the invoice picture, obtaining each pending area; according to each pending area and invoice recognition model, the identification result of each to be identified is obtained; Wherein, the invoice recognition model is obtained based on invoice training samples and corresponding tag information training; if it is determined that the identification result of each of the respective peers to be identified is verified by the identification result check, the employee registration information is obtained according to the employee ID; according to the invoice Information, the employee registration information, and the reimbursement single mode, generated an invoice reimbursement order; wherein the invoice information is obtained from the identification results of each pending area.

[0130] Those skilled in the art will appreciate that embodiments of the present invention can be provided as a method, system, or computer program product. Accordingly, the present invention can be in the form of a fully hardware embodiment, a full software embodiment, or a combination of software and hardware aspects. Moreover, the present invention can be in the form of a computer program product implemented in one or more computers that include a computer available program code (including, but not limited to, disk memory, CD-ROM, optical memory, etc.).

[0131] The present invention is described with reference to a flowchart and / or block diagram of a method, device (system), and computer program product, in accordance with an embodiment of the present invention. It should be understood that each of the flowcharts and / or blocks in the flowchart and / or block diagram can be implemented by a computer program command, and a combination of flow and / or box in the flowchart and / or block diagram. These computer program instructions can be provided to a general purpose computer, a dedicated computer, an embedded processor, or another programmable data processing device to generate a machine such that instructions executed by the processor of the computer or other programmable data processing device. Implementation in the process Figure one Process or multiple processes and / or boxesFigure one Apparatus specified in multiple boxes or multiple boxes.

[0132] These computer program instructions can also be stored in a computer readable memory capable of booting a computer or other programmable data processing device in a particular manner, making the instructions stored in the computer readable memory generate a manufacturing article of instruction devices, whichDevice is implemented in the process Figure one Process or multiple processes and / or boxes Figure one The function specified in the box or multiple boxes.

[0133] These computer program instructions can also be loaded on a computer or other programmable data processing device such that a series of steps are performed on a computer or other programmable device to generate a computer implemented process, thereby executing on a computer or other programmable device.The instruction is provided for implementation Figure one Process or multiple processes and / or boxes Figure one The steps of the function specified in multiple boxes or multiple boxes.