单页PDF转UIImage后文本识别质量不佳的优化方案求助

各位大佬好，这个问题我卡了好久了，实在没辙了来求帮忙！

我现在要处理用户的PDF格式每日任务文档，核心是要从里面提取最多20组「日期+机场/时间」的组合。原本直接提取PDF文本的时候，内容的顺序完全乱套，日期和对应的机场/时间根本对不上，试了.string和.attributedString两种方式都解决不了顺序问题。

后来我改成了把PDF转成UIImage，再用Vision框架提取带排版结构的文本——这个思路逻辑上是对的，能保证内容顺序和原PDF一致，但新问题来了：PDF转图片的过程中质量损失太严重，要么漏识别文本，要么直接识别错误，完全没法稳定工作。肯定有更稳健的PDF转图+文本提取的方法吧？

我先贴一下目前用的PDF转UIImage的代码：

func drawPDFfromURL(url: URL) -> UIImage? {
    guard let document = PDFDocument(url: url), let page = document.page(at: 0) else { return nil }
    
    let scale = CGFloat(3.0)
    let pageRect = page.bounds(for: .mediaBox)
    let scaledSize = CGSize(width: pageRect.width * scale, height: pageRect.height * scale)
    
    let rendererFormat = UIGraphicsImageRendererFormat()
    rendererFormat.scale = 1 // 也试过用UIScreen.main.scale
    rendererFormat.opaque = false
    
    let renderer = UIGraphicsImageRenderer(size: scaledSize, format: rendererFormat)
    
    let image = renderer.image { ctx in
        let context = ctx.cgContext
        UIColor.lightText.set() // 设成lightText解决了不少问题
        context.fill(CGRect(origin: .zero, size: scaledSize))
        
        context.saveGState()
        // 翻转和缩放以正确渲染
        context.translateBy(x: 0.0, y: scaledSize.height)
        context.scaleBy(x: scale, y: -scale)
        
        // 尝试增强细字体的线条宽度
        context.setLineWidth(2.0)
        context.setLineJoin(.round)
        context.setLineCap(.round)
        
        page.draw(with: .mediaBox, to: context)
        context.restoreGState()
    }
    
    return image
}

再贴一下Vision文本提取的代码：

func getTextFromImage(image: UIImage) {
    guard let cgImage = image.cgImage else { return }
    
    let request = VNRecognizeTextRequest { request, error in
        guard let results = request.results as? [VNRecognizedTextObservation] else {
            print("No text found.")
            return
        }
        
        var wordItems: [(text: String, rect: CGRect)] = []
        for observation in results {
            guard let candidate = observation.topCandidates(1).first else { continue }
            let stringRange = candidate.string.startIndex..<candidate.string.endIndex
            
            if let boxObservation = try? candidate.boundingBox(for: stringRange) {
                let normBox = boxObservation.boundingBox
                let pixelBox = VNImageRectForNormalizedRect(
                    normBox, Int(image.size.width), Int(image.size.height)
                )
                wordItems.append((candidate.string, pixelBox))
            }
        }
        
        // 按Y坐标分组（行逻辑），Y容差控制每行的匹配范围
        let yTolerance: CGFloat = 35.0
        let grouped = Dictionary(grouping: wordItems) { word in
            round(word.rect.origin.y / yTolerance) * yTolerance
        }
        
        // 从上到下排序行
        let sortedLines = grouped.sorted(by: { $0.key > $1.key })
        
        // 每行内从左到右排序单词并拼接
        let lines = sortedLines.map { (_, words) in
            words.sorted(by: { $0.rect.origin.x < $1.rect.origin.x })
                .map(\.text)
                .joined(separator: " ")
        }
        
        let rawText = lines.joined(separator: "\n")
        print("OCR with Layout:\n\(rawText)")
    }
    
    request.recognitionLevel = .accurate
    request.recognitionLanguages = ["en-US"]
    
    let handler = VNImageRequestHandler(cgImage: cgImage, options: [:])
    try? handler.perform([request])
}

我自己也试过调整不少参数：比如UIColor.lightText.set()确实解决了不少显示问题，也试过改缩放比例、yTolerance、setLineWidth这些，但始终没法解决图片质量导致的识别误差问题。有没有更靠谱的PDF转图+文本提取的方案？求各位大佬指点迷津！

内容来源于stack exchange