基于LP-CAAPPCNN的低亮度苹果图像增强研究

A Study on Low-Light Apple Image Enhancement Using LP-CAAPPCNN

摘要: 针对低亮度环境下采集的苹果图像存在亮度低、对比度差、细节丢失及噪声放大等问题，本文提出一种基于LP-CAAPPCNN的图像增强方法。该方法首先通过拉普拉斯金字塔将亮度图像分解为多个高频细节层与一个低频近似层，实现噪声与结构信息的分离；在此基础上，设计上下文感知的像素级自适应参数机制，依据局部亮度与纹理统计动态计算脉冲耦合神经网络的链接强度、衰减系数及阈值幅度，并对低频近似层进行精准增强；最后，通过引导滤波后处理，实现噪声抑制与视觉质量的联合优化。在低亮度苹果图像数据集上进行验证，实验结果表明实验结果表明，本文所提方法在信息熵等客观指标上均优于对比方法，能够有效提升低亮度苹果图像中的缺陷识别能力，并提升高质量检测水平。

Abstract: To address issues such as low brightness, poor contrast, loss of detail and noise amplification in images of apples captured in low-light environments, this paper proposes an image enhancement method based on LP-CAAPPCNN. This method first decomposes the luminance image into multiple high-frequency detail layers and a low-frequency approximation layer using a Laplacian pyramid, thereby separating noise from structural information. Building on this, a context-aware, pixel-level adaptive parameter mechanism is designed to dynamically calculate the connection weights, decay coefficients and threshold amplitudes of the spiking-coupled neural network based on local luminance and texture statistics, enabling precise enhancement of the low-frequency approximation layer; Finally, through post-processing via guided filtering, noise suppression and visual quality are jointly optimised. Validation on a dataset of low-luminance apple images demonstrates that the proposed method outperforms comparison methods in objective metrics such as information entropy, effectively enhancing defect recognition capabilities in low-luminance apple images and improving the quality of detection.