You mention convolution which implies that your arrays represent time domain signals. As this is digital signal processing, the time is discretized and every element of the array corresponds to a sample. Two consecutive samples have a fixed time distance, say $T$.
If you convolve the discrete time signal with the impulse response of a low-pass filter (LPF) the resulting ouput signal will be longer than the input signal, because in general, linear time-invariant (LTI) systems have a memory (length of impulse response). You cannot reduce the array length without distorting the output signal (unless you use block-wise processing with overlap-add/-save, but that's another story). So, no matter what type of filter the impulse response of your LTI system represents, you cannot just remove elements from the output array.
To find out which frequency components are attenuated by a system you should analyze its transfer function (sometimes also refered to as frequency response). The transfer function is obtained by transforming the impulse reponse into the frequency domain. A tool for doing that with DSP is the discrete Fourier transform (DFT) which almost always comes in form of its FFT implementation (Fast Fourier transform). In case that the system in question is a black box and its impulse response is unknown the frequency domain representation of input and output signal can be compared to find out what frequencies are attenuated or amplified.