**1. Introduction**

Photoacoustic (PA) imaging, an emerging biomedical imaging modality based on PA effect, has been developed extensively in recent years. During PA imaging, laser pulses energy are delivered into biological tissues, leading to transient thermoelastic expansion and thus wideband ultrasound emission [1]. The generated ultrasound waves propagate to the surface where detected by ultrasound transducers [2], and are reconstructed into an image. As PA images can reveal pathology features and physiological structures according to the specific optical absorption distribution of biological tissues, the diagnoses on the tissue differs in physiological properties a lot, such as breast cancer diagnosis [3], and hemodynamics monitoring [4,5], making PA imaging a promising and high potential imaging modality.

Since typical detected PA signals of ideal optical absorbing particle is a single bipolar N-shape pulse [6,7], PA signals of a complicated biological tissue can be considered to be the combination of individual N-shape pulses. However, the N-shape wave basis results in two problems: the first problem is aliasing between adjacent micro-structures. The signal of a tiny target can be affected and even buried by the bipolar signal of a large target at a short distance, leading to unexpected aliasing and distortion in the final image. The second problem is that the existence of the N-shape wave complicates subsequent imaging work. When reconstructing images, the envelope of signal must transform the negative part into positive, as the negative part possesses significant information as well. Both drawbacks could deteriorate the quality of PA images.

Therefore, the processing on an N-shape wave can be significant in alleviating the drawbacks and has been investigated a number of studies. Li [7], identified some properties about the N-shape wave and introduced processing methods, including wavelet and deconvolution. Ermilov [8,9], transformed the bipolar N-shape pressure pulse to the monopolar pulse using wavelet transform in order to ge<sup>t</sup> the signal suitable for tomographic reconstruction. The deconvolution method can convert the bipolar N-shape wave to the monopolar wave and is rarely applied to processing signals, and thereby raising our interest. Traditionally, the deconvolution method is often applied to image reconstruction and image processing. For instance, Kruger [10] and Gamelin [11] used deconvolution methods in PA image reconstruction. Cai [12] improved the image resolution of PA microscopy by using deconvolution method on images. Recently, Nagaoka [13] proposed a reconstruction method to improve axial resolution through the suppression of the time side lobes in PA tomography by Wiener filtering.

In this paper, we propose using the image optimization method to process raw PA signals using signal deconvolution and empirical mode decomposition (EMD). This method mainly relies on signal deconvolution. However, as unexpected artifacts appeared around the imaging target only with signal deconvolution, EMD was instead adopted into our method as a subsequent step to further improve the reconstructed image quality. Moreover, we averaged multi-sampling raw PA signals with time shift correction by cross-correlation method as a preprocessing step in order to raise the signal-to-noise ratio (SNR) and enhance the performance of signal deconvolution.
