Image Recognition

No, you will also need various sensors to "feed" data to the PLC. But as far as the "brains" are concerned, I believe that a suitable PLC can do it, whatever "it" is. However, there might be other ways ranging from a dedicated microprocessor to a general-purpose PC.

Tornado is correct in pointing out that you'd need other sensors and such.

You'd probably need or want some form of machine vision device. In general, these need more calculating power than the typical (current) PLC has. I haven't looked into the field recently--I'm guessing some machine vision devices come with the built in smarts needed. Otherwise, they need some sort of "host" computer more powerful than a typical (current) PLC.

(I should point out that my knowledge of current PLCs is probably 10 years out-of-date.)

I have been working on this for the last 5 years.. I would be interested in what Ideas (if any) you have to enable recognition and stereoscopic vision..

Right. Object recognition beyond the basic "symbol recognition" (road-signs, graphic symbols and letter recognition - as seen in OCR applications) is far more complicated than first thought some 20 years ago, because real-world objects have an actual volume, seen different from every angle as it appears on camera, and for software to analyse the camera input, it needs a monster database just to be able to sieve out one given rigid, stationary object - apart from the background - not to mention an assortment of "soft" object - such as a face for instance.

As for facial recognition specifically - a good deal of advance was made in Germany in the last five years - using proportion-measurement and comparison algorythms.

Whatever was thought to have required simple comparison techniques, is now thought as a good candidate to Artificial Intelligence no-less...

The reason is just about similar to that of natural language translation, and real-world reasoning in general - understanding in the real sense of the word, requires a world of prior knowledge to be drawn from and compared to, and this means a near-impossible calculation sequence of a monster database, often with conflicting notions built-in.

Such task was encountered using "Deep Blue" vs Casparov in Chess, and it was accomplished using Heuristic-type "educated" guesswork, because the database required to mark all possible future chess moves from a given position is too big to write, let alone sieve through - "scan" as it may...

This is not my field, but I don't think you can get the kind of optical recognition and force control feedback you need to both find and pick veggies without damage. I think you'll need more computational power.

Interesting idea.

Just by learning PLC course you cannot design a robot machine for any applications. It needs sound knowledge in hardware and software. Knowledge of PLC will assist you to design the control circuit. Besides such type of innovative projects needs tam work and contribution from various specialists.

What kind of vegetables and where to pick from?

We start with Caulliflower..

then other ground based veg..

Have you seen this strawberry picking robot..?

I still prefer them hand picked.

What? I'm old fashioned that way..

Yes, you can actually.

My BSECE colleagues actually pursue a project of image recognition way back college. However, talking about PLC, most application is for industrial controls, but your idea will work certainly. PLC accept analog and digital outputs and the same time controls an output. Commonly it is designed and commercialized that way. Therefore, if you have to make it a PLC based, you need additional Image recognition sensor or receptor system to command the PLC 1 or 0. It could be briefly integrated this way.

Image recognition system <----> PLC <---->inputs & outputs

By the way, my colleagues made a successful project way back then. Pretty sure you can make it also.

Good luck!

Not sure what you are actually asking. Are you planning on building (making) a robot to pick or are you going to but a robot and program it (make) to pick?

If it is the former then the answer is no.

If it is the latter then the answer, as with many other things in life is, "it depends". It depends on the programming environment of the robot controller - for example Fanuc have quite an easy programming environment whereas say Adept may be deemed to be slightly more complicated. As Tornado pointed out you will need cameras etc to interface to the system. Most robots come with their own control system and have alot of the necessary built in features/interfaces so a stand alone PLC may not be required at all. As a matter of interest most of the chocolate boxs with assorted chocolates are picked and packed by robots with a vision system - for example if you look at SIG PAC (based in Switzerland) - there are others aswell.

I have been working with FANUC on this project .. they have some excelent robots..

The PLC is the brains, so the interface between the brain are the sensors like vision systems, color detectors, robot arms and sensors to detect the pressure on the item as well as the "feel" of it.

Once you have the input sensors and output actuators, the brain has to be programmed to work with them not to damage the crop.

Picking vegetables...

With a few exceptions, vegetables are harvested stems, stalks, roots and all. Usually some sort of mechanical separation, then various parts are directed to the proper discharge chute. Efficient equipment for these processes already exist.

Why would you use a programmable robot to pick cabbages? They are grown in rows and commercial varieties will all be of a similar size when picked.

Mount a side gripping conveyor at 30-45° with the leading end 10-20mm above the soil level. As the conveyor rig is pulled along the row by a tractor, the cabbages will feed into the conveyor and the angle will pull them out of the ground. As the cabbage lifts, pass it over a blade to cut off the root ball which drops into a shredder and deposits it as mulch on the field. Next pass it under sensor that determines the diameter of the head and instructs a pair of cutters with concave blades to how far to come in to cut off the outer leaves, leaving a bear minimum for display purposes in the retail outlet. If the leaf cutters have a box motion the conveyor can move continuously. The outer leaves would be sucked/blown into the shredder. Finally a blade mounted above the conveyor cuts the head away from the short length of stem held in the conveyor. The stem also goes into the shredder. An arm sweeps the head into a form fill bagger which evacuates the bag, replaces a small amount of inert gas to increase shelf life and hermetically heat seals the bag. The packaged bag drops onto a scale, a label is printed and attached and the finished product is deposited into a box for shipping.

The process is totally automated, picks and packs in less than 30 seconds (good marking point) and is cheap to engineer. It does not need to be programmable, you do not need a robot.

There is an unfortunate flaw in your proposal.. a field of cauliflower will not all reach the same size.. at the same time.. Commercial growers in the UK will have to visit a Field a minimum or three times and up to seven times is not uncommon..in order to get a consistent crop for the supermarket buyers..This is why I have been working on the problem for the last five years..

I bow to your superior knowledge of cabbage growing. You have been working on this for five years, me for 5 minutes.

So the conveyor becomes a pair of synchronised 2"-3" link chains running each side of the row. Each link is fitted with a spring loaded arm that moves in and grips the cabbage when de-latched. (Spring loaded plungers with bellow to keep the dirt out might be robust and simple enough to work. I would avoid any sort of scissor action which I imagine would give reliability problems in these conditions) Reposition the sensor so that it determines the head size as the cabbage passes between the chains at their lowest point. If the head is the correct size, trigger a solenoid to de-latch the gripper arms. You can now pick cabbages selectively.The subsequent lift, cut and pack actions are as before. Store the head size in a shift register for the leaf cutters. Reset the gripper arms with a cam on the return leg of the conveyor.

I am not being paid to design this thing, I am just throwing in ideas to justify my main premise, that a robot is overkill for this application.

I would also make the point that farmers are a very conservative (small c) market, which I agree is changing, but very slowly. Getting industry to adopt robots to move regular sized square boxes in a nice clean factory is hard enough. Putting one in a muddy field in a maintenance environment where the norm is "if I can weld it, I can fix it", is going to be a nightmare.

This sounds like a job for an optical sorter, which is what a pack shed would use

Not exactly portable though

Sorter's have controlled conditions for the cameras & other sensors. Taking that to the field is not as easy as it sounds.

Being able to identify which cauliflower is ripe is only the 1st step

The harvester 2nd

The integration of the 2 in a package that will roll down the rows is the future...

cool project

Simple optical is probably not the right way to go here. as there are too many variables. Combinations of sensors, typically a microwave, IR or UV that recognises the difference between the water content of leaves and florets, combined with an ultrasonic for size, and an optical for colour. What you are trying to simulate is a human look, feel, and make an instant judgement, evaluation process.

Once the selection has been made, the lifting and processing aspects are not very dissimilar to lots of other semi automated harvesting machines. I may have been stretching it a bit to pack in the field, but I think the supermarkets would kneel before you and beg to sell that degree of freshness to their punters.

An Optical Sorter is quite a bit more than a camera, depending on the product

it's gonna be tough to beat the standard harvester Merc described :D

Oh I went the Ultrasonic route.. and I found I got no definition without contact and no definition with contact because of the air and rain water pockets that are completely random in the leaf structure.. The water content of the plant is not as far as I can tell consistent and can depend on the last rainfall drainage of the field sunlight and wind and will vary from plant to plant.. Microwave at 4Ghz gives reasonable definition but of all the wrong things the stalks and the leaf on the surface without enough penetration to be able to size the Curd..2 Ghz gave reasonable penetration but lousy definition, so you know its there but where exactly... So bring the wall back over here I need to give myself another headache..

Optical sorter in the pack house might be a possibility.. however the first step is to identify and pick the cauliflower head itself.. the CURD that is the white bit we want to consume is covered with a dense layer of leaves.. so we have a green ball in a sea of other green balls also there is a progressive hight differential south-north in the Field like a wave.. but you have to stand in the middle of the plants to see it.. It is caused (as far as I can tell) by the sun -shadow effect on the plants so some get more sun they cast shadow on ones behind and stunt their growth slightly. The fact is that at the moment a biped with stereoscopic vision and a degree of logic is the best way to harvest the crop .. but. When paid by the units cut some operatives will cut smaller than the minimum required size and jeopardies the whole days work by putting small ones into the trays. Also it is a fact that if you cut a small cauli then you have to throw it away in Field because there is no use for it. so that is money down the hole in the ground.

Generally I have tried cameras.. they offer no differential that is good enough to identify the cauliflower head, I tried Infra red ..again no differential..I have tried laser, too much information. I thought about Xray.. but the health issues are a bit scary. I went to T-waves..Spectacular costs and not portable enough. So if anybody has a bright Idea that I didn't try I would like to hear about it..

I think you should take on board what jhhassociates is saying. Pick and place robots do work well in industry however as they have more features say cameras, sensors etc then the working environment becomes an issue. I have seen ABB robots working in cement factories where there are covered in dust and grit but they did not have any cameras that they needed to worry about. I have also worked with robots in the food and pharma industry and these environments are much cleaner as you can imagine - however the slighest issue with the vision system required intervention, I just can't imagine farmer X sitting down and reprogramming the unit or refocusing the vision system . Is it your intention that the robot would be out in the fields working?

Best of luck with it anyway.

The crop is asparagus. Asparagus grows constantly, and is cut everyday, so there's different sizes in the field, and only the ones over 6 inches are to be picked. The ground is pritty flate and level, but sometimes there's rye in field, it's green and as tall as the asparague (it's to block the wind from blowing the sand which scars the asparagus). The asparagus is to be cut just below ground level. The most tricky part is that sometimes the spears come right up beside eachother, and a long one to be picked will be beside a short one, or surrounded by short ones. In this case the short ones often get cut or dammaged by the pickers anyways, so some fallout is ok. The spears come up in rows about 1.5 feet wide, and they randomly occur in this area.

OK so that is a "above ground -below surface " situation..you need a combination of sensors.. the green on green will not help..

I think you will get a result there if you go down the density path...

how about this image recognition. it is the latest optical technology. i have uesed it for years. doing a great job on recognize images and pictures.

Unfortunately field grown ground plants are not flat images and do not normally contain any other colour but GREEN. Green on Green on Green as any Image analyst will tell you is very difficult to differentiate. I looked at the link you offered but was not able to find anything that would improve my present position. If you can please elaborate on what you think would be of help