Despite the progress made in the biomechanics of martial arts for the past three decades, there is still much work that needs to be done in order to improve the understand the complexity of these ancient arts (Neto et al., 2008b). In martial arts and contact sports strikes are performed with and without impact at near maximum hand speeds. For that reason, electromyography (EMG) analysis of such movements is non-trivial. Studies about EMG of martial arts strikes are rare. Neto et al. (2008a) reported a kinematical and EMG analysis of Kung Fu Yau-Man Palm strikes without impact. An empirical model applied to data obtained by a high-speed camera (1000 Hz) described the kinematical characteristics of the movement. The authors analyzed the EMG signals of the triceps brachii (TB), biceps brachii (BB) and brachioradialis (BR) muscles during the strikes in the time and frequency domains. The EMG results showed a well developed muscle coordination of the practitioners in agreement with kinematical results. In a similar study, Neto et al., 2007b evaluated the coordination between agonists and antagonists muscles of the arm in a movement of Kung Fu performed by trained and untrained subjects. The authors demonstrated, using wavelet transforms to analyze the data in the frequency domain through time, that the untrained subjects presented much higher undesired co-contractions during the strikes. Neto et al., 2007a compared the EMG activity of the TB, BB and BR during palm strikes with and without impacts. EMG analyses were done in the time and wavelet domains. Morlet wavelet power spectra were obtained and an original method was used to quantify statistically significant regions on the power spectra (SSP). The results both in the time (root mean square - rms) and frequency domains (SSP) indicated higher TB and BR muscle activity for the strikes with impacts. No significant difference was found for the BB in the two different scenarios.

This paper has three main goals. Firstly, to investigate the differences in the EMG activity of muscles during strikes performed with and without impacts. Secondly, to confirm some of the advantages, reported by Neto et al., 2007a, of using SSP values instead of rms values when analyzing EMG data. Finally, to introduce a new method of calculating median frequency values (MDF) using wavelet transforms.

Eight KF Yau-Man practitioners with 4.5 year average training experience of were selected to participate in the experiment. Each participant performed one palm strike without impact and one strike with impact targeting a training shield held by their KF instructor. A detailed description of the palm strike movement can be found on Neto et al., 2007c. Surface EMG signals were obtained from the anterior deltoid (DA), TB and BR of their striking hand. The subjects were allowed to position themselves in relation to the target and to adjust its height as they wished.

The methodology was approved by the University of Vale do Paraiba Ethics in Research Committee (Protocol #: H226/2007/CEP) and the subject provided his informed written consent.

Materials
The surface EMG signals were obtained using a four-channel module (model EMG400C, EMGSystem, Brazil) with a total amplifier gain of 2000 and sampled at 1000 Hz. A 12 bits AD converter digitalized the analogue signals with a sampling frequency of anti- aliasing of 2.0 kHz for each channel and an input range of 5mV. After shaving and cleaning the skin with alcohol, bipolar surface EMG electrodes were placed according to standard procedures with an inter-electrode distance 0.2 cm (Hermens and Freriks, 2000) and guided by bone prominences and the route of the muscle fibers.

Data analyses
All EMG data was processed off-line with Matlab 7.0.1 (MathWorks Inc). The EMG signals were treated in two different ways. First, the EMG signals were analysed following standard procedures in the time and frequency domains. For the time domain analysis the EMG signals were full wave rectified and then linear smoothed using a low pass Butterworth order 4 filter with a cut frequency of 14 Hz. The mean amplitude for this linear envelope was determined and it was used to normalize the EMG signals from which rms values were calculated. For the frequency domain, the median frequency of the signals were found using Fourier Transform (FMDF).

Second, EMG signals were normalized by the standard deviation and treated using Morlet wavelet transform as in Neto et al., 2007a using a base algorithm developed by Torrence and Compo, 1998. Sum of the significant power values (SSP) on the wavelet power spectra were calculated as in Neto et al., 2007a, and an algorithm was developed in Matlab 7.0.1 (MathWorks Inc.) to determine the MDF associated to the wavelet significant power values (WMDF).

Statistics
A balanced analysis of variance (ANOVA) was performed for the variables rms, FMDF, SSP and WMDF with one random factor (Subject) and two different crossed fixed factors (Impact and Muscle), followed by dependent t tests with Bonferroni corrections post hoc analyses. Pearson's Correlation analysis was used to investigate possible linear associations between the variables. Inter-subject (IECV) coefficients of variation for each muscle with and without impact were also calculated for each variable. All statistical tests were done using the Minitab (MINITAB^® 14.12.0, Minitab Inc.) software; values of p smaller than 0.05 were considered significant.

The least squares mean for rms was 0.23 for the strikes with impact and 0.21 for the strikes without impacts. ANOVA for the rms demonstrated that only the factor Subject as significant (Table 1).

The least squares mean for FMDF was 126.5 Hz for the strikes with impact and 124.1 Hz for the strikes without impacts. ANOVA for the FMDF demonstrated no significant factors (Table 2).

The least squares mean for SSP was 31931 V² for the strikes with impact and 22750 V² for the strikes without impacts. ANOVA for the SSP demonstrated only the factor Impact as significant (Table 3). Mean SSP was significantly higher (p = 0.023) for the strikes with impact (31931 ± 18153 V²) than for the strikes without impacts (22750 ± 4987 V²).

The least squares mean for WMDF was 80.5 Hz for the strikes with impact and 93.5 Hz for the strikes without impacts. ANOVA for the WMDF demonstrated that all three factors, Subject, Impact and Muscle were significant (p = 0.022; p = 0.007; p = 0.005, respectively) (Table 4). Mean WMDF was significantly lower (p = 0.007) for the strikes with impact (80.5 ± 24.3 Hz) than for the strikes without impacts (93.5 ± 12.7 Hz).

No significant linear association was found between the values of rms and SSP (R = -0.028, p = 0.849) and FMDF and WMDF (R = 0.270, p = 0.063).

Table 5 shows the rms, FMDF, SSP and WMDF mean values and IECV values for each muscle with and without impact. The mean and coefficient of variations (in percentage) considering all data for the values of rms, SSP, FMDF and WMDF were: 0.22 (56%); 125.3 Hz (19%); 27340 V2 (51%); and 87.03 Hz (23.3%), respectively.

• The results show higher muscle activity and lower electromyography median frequencies for strikes with impact compared to strikes without.
• SSP results presented higher sensitivity and lower inter-subject coefficient of variations than rms results.
• Kung Fu palm strikes with impact may present better motor units' synchronization than strikes without.