Azure Document Intelligence ile .NET'te Form ve Fatura Analizi

Azure Document Intelligence ile .NET'te Form ve Fatura Analizi

Azure Document Intelligence (eski adıyla Form Recognizer), belgelerden yapılandırılmış veri çıkaran bir AI hizmetidir. Faturalar, sözleşmeler, kimlik belgeleri ve özel formlardan anahtar-değer çiftleri, tablolar ve metin bloklarını otomatik olarak tanıyabilir.

Kurulum

dotnet add package Azure.AI.FormRecognizer

Fatura Analizi (Prebuilt Invoice Modeli)

using Azure;
using Azure.AI.FormRecognizer.DocumentAnalysis;

public class InvoiceAnalysisService
{
    private readonly DocumentAnalysisClient _client;

    public InvoiceAnalysisService(IConfiguration config)
    {
        _client = new DocumentAnalysisClient(
            new Uri(config["DocumentIntelligence:Endpoint"]!),
            new AzureKeyCredential(config["DocumentIntelligence:Key"]!));
    }

    public async Task<InvoiceData> AnalyzeInvoiceAsync(Uri invoiceUri)
    {
        var operation = await _client.AnalyzeDocumentFromUriAsync(
            WaitUntil.Completed,
            "prebuilt-invoice",
            invoiceUri);

        var result = operation.Value;
        var invoice = result.Documents[0];

        return new InvoiceData
        {
            VendorName    = GetFieldValue(invoice, "VendorName"),
            CustomerName  = GetFieldValue(invoice, "CustomerName"),
            InvoiceId     = GetFieldValue(invoice, "InvoiceId"),
            InvoiceDate   = GetDateField(invoice, "InvoiceDate"),
            DueDate       = GetDateField(invoice, "DueDate"),
            SubTotal      = GetCurrencyField(invoice, "SubTotal"),
            TotalTax      = GetCurrencyField(invoice, "TotalTax"),
            InvoiceTotal  = GetCurrencyField(invoice, "InvoiceTotal"),
            LineItems     = ExtractLineItems(invoice)
        };
    }

    private string? GetFieldValue(AnalyzedDocument doc, string fieldName)
    {
        if (doc.Fields.TryGetValue(fieldName, out var field))
            return field.Content;
        return null;
    }

    private DateTimeOffset? GetDateField(AnalyzedDocument doc, string fieldName)
    {
        if (doc.Fields.TryGetValue(fieldName, out var field) && field.FieldType == DocumentFieldType.Date)
            return field.Value.AsDate();
        return null;
    }

    private double? GetCurrencyField(AnalyzedDocument doc, string fieldName)
    {
        if (doc.Fields.TryGetValue(fieldName, out var field) && field.FieldType == DocumentFieldType.Currency)
            return field.Value.AsCurrency().Amount;
        return null;
    }

    private List<InvoiceLineItem> ExtractLineItems(AnalyzedDocument doc)
    {
        var items = new List<InvoiceLineItem>();
        if (!doc.Fields.TryGetValue("Items", out var itemsField)) return items;

        foreach (var item in itemsField.Value.AsList())
        {
            var fields = item.Value.AsDictionary();
            items.Add(new InvoiceLineItem
            {
                Description = fields.TryGetValue("Description", out var desc) ? desc.Content : null,
                Quantity    = fields.TryGetValue("Quantity", out var qty) ? qty.Value.AsDouble() : null,
                UnitPrice   = fields.TryGetValue("UnitPrice", out var price) ? price.Value.AsCurrency().Amount : null,
                Amount      = fields.TryGetValue("Amount", out var amt) ? amt.Value.AsCurrency().Amount : null,
            });
        }

        return items;
    }
}

public record InvoiceData(
    string? VendorName, string? CustomerName, string? InvoiceId,
    DateTimeOffset? InvoiceDate, DateTimeOffset? DueDate,
    double? SubTotal, double? TotalTax, double? InvoiceTotal,
    List<InvoiceLineItem> LineItems);

public record InvoiceLineItem(
    string? Description, double? Quantity, double? UnitPrice, double? Amount);

Kimlik Belgesi Analizi

public async Task<IdDocumentData> AnalyzeIdDocumentAsync(Stream documentStream)
{
    var operation = await _client.AnalyzeDocumentAsync(
        WaitUntil.Completed,
        "prebuilt-idDocument",
        documentStream,
        new AnalyzeDocumentOptions { Locale = "tr-TR" });

    var doc = operation.Value.Documents[0];

    return new IdDocumentData
    {
        FirstName   = GetFieldValue(doc, "FirstName"),
        LastName    = GetFieldValue(doc, "LastName"),
        DocumentNumber = GetFieldValue(doc, "DocumentNumber"),
        DateOfBirth = GetDateField(doc, "DateOfBirth"),
        Nationality = GetFieldValue(doc, "Nationality"),
        Confidence  = doc.Confidence
    };
}

Özel Model Eğitimi

Kendi form şablonlarınız için özel model eğitebilirsiniz:

public class CustomModelTrainer
{
    private readonly DocumentModelAdministrationClient _adminClient;

    public async Task<string> TrainCustomModelAsync(
        Uri trainingDataUri,
        string modelId,
        string description)
    {
        var operation = await _adminClient.BuildDocumentModelAsync(
            WaitUntil.Completed,
            modelId,
            DocumentBuildMode.Template, // veya Neural
            trainingDataUri,
            new BuildDocumentModelOptions
            {
                Description = description
            });

        Console.WriteLine($"Model eğitildi: {operation.Value.ModelId}");
        Console.WriteLine($"Desteklenen belge türleri: " +
            $"{string.Join(", ", operation.Value.DocumentTypes.Keys)}");

        return operation.Value.ModelId;
    }

    public async Task<AnalyzeResult> AnalyzeWithCustomModelAsync(
        string modelId, Stream documentStream)
    {
        var operation = await _client.AnalyzeDocumentAsync(
            WaitUntil.Completed,
            modelId,
            documentStream);

        return operation.Value;
    }
}

Toplu Belge İşleme

public class BatchDocumentProcessor
{
    private readonly InvoiceAnalysisService _analyzer;
    private const int MaxConcurrentRequests = 5;

    public async Task<IReadOnlyList<ProcessingResult>> ProcessBatchAsync(
        IEnumerable<string> documentUrls,
        CancellationToken ct = default)
    {
        var semaphore = new SemaphoreSlim(MaxConcurrentRequests);
        var results = new ConcurrentBag<ProcessingResult>();

        await Parallel.ForEachAsync(documentUrls,
            new ParallelOptions { MaxDegreeOfParallelism = MaxConcurrentRequests, CancellationToken = ct },
            async (url, token) =>
            {
                await semaphore.WaitAsync(token);
                try
                {
                    var data = await _analyzer.AnalyzeInvoiceAsync(new Uri(url));
                    results.Add(new ProcessingResult(url, data, Success: true));
                }
                catch (Exception ex)
                {
                    results.Add(new ProcessingResult(url, null, Success: false, Error: ex.Message));
                }
                finally
                {
                    semaphore.Release();
                }
            });

        return results.ToList();
    }
}

ERP Entegrasyonu

[HttpPost("invoices/process")]
public async Task<IActionResult> ProcessInvoice(IFormFile file)
{
    await using var stream = file.OpenReadStream();

    var operation = await _client.AnalyzeDocumentAsync(
        WaitUntil.Completed, "prebuilt-invoice", stream);

    var invoiceData = _mapper.MapToInvoiceDto(operation.Value.Documents[0]);

    // ERP sistemine aktar
    var erpResult = await _erpService.CreateInvoiceAsync(invoiceData);

    return Ok(new { InvoiceId = erpResult.Id, Status = "Created" });
}

Desteklenen Prebuilt Modeller

Model	Kullanım Amacı
prebuilt-invoice	Faturalar
prebuilt-receipt	Perakende fişleri
prebuilt-idDocument	Kimlik belgesi, pasaport
prebuilt-businessCard	Kartvizitler
prebuilt-contract	Sözleşmeler
prebuilt-read	Genel OCR
prebuilt-layout	Tablo ve yapı analizi

Sonuç

Azure Document Intelligence, belge işleme süreçlerini otomatikleştirmenin en etkili yollarından biridir. Muhasebe, lojistik ve finans sektöründe fatura, irsaliye ve sözleşme işleme sürelerini dramatik biçimde kısaltabilir. Özel modeller sayesinde kendi şablonlarınıza özel çıkarım yapabilirsiniz.