Can this be considered as a trend? I have this time series with a trend line plotted in R using the lowess function, f=0.2 (I couldn't use stl or decompose function because this data is measured annually). Although this seems to fit the data in some way, I'm hesitant to conclude this as a "trend" because trend is defined as a "long run increase/decrease over time"; but this here has fluctuations.
 A: Yes, I would define this LOESS fit as a trend.
Your definition of the trend is predicated on "long run" and defining what you mean by "long run". You could fit a straight line through those data and it would likely have a slightly positive slope, but it would be a very poor description of the data to hand, but it would also be a justifiable description of the overall magnitude of the change in Y over the observed series.
Whether you find that linear trend line useful depends on the question you have in mind. And you will likely want to take account of the glaring lack of independence in the residuals — arising from the unmodelled decadal-scale variation in the response — if you wanted to calculate an uncertainty estimate for this long term trend.
You may not find this useful however as a definition of the trend and therein lies the rub; "long run" is whatever you want to define it as and that will depend on the question you are asking. If I want to know if there are trends in a population of bacteria in response to some environmental peturbation, I likely won't want to define "long run" as many decades; minutes would probably be more appropriate.
I prefer a looser definition, one where a trend is defined simply as a change over time in the underlying mean or level of a time series.

You could instead fit these data using a generalised additive model (GAM) and get a more objective way of choosing the non-linear trend. If you want more on this aspect, I have written some blog posts on the topic: https://fromthebottomoftheheap.net/2016/04/10/loess-revisited/ and https://fromthebottomoftheheap.net/2012/07/24/whats-wrong-with-loess-for-palaeo-data/ for example, which should have links to other posts on GAMs if you're unfamiliar with those models.
